2023-01-05 20:25:55 +08:00
/**************************************************************************/
/* bindings_generator.cpp */
/**************************************************************************/
/* This file is part of: */
/* GODOT ENGINE */
/* https://godotengine.org */
/**************************************************************************/
/* Copyright (c) 2014-present Godot Engine contributors (see AUTHORS.md). */
/* Copyright (c) 2007-2014 Juan Linietsky, Ariel Manzur. */
/* */
/* Permission is hereby granted, free of charge, to any person obtaining */
/* a copy of this software and associated documentation files (the */
/* "Software"), to deal in the Software without restriction, including */
/* without limitation the rights to use, copy, modify, merge, publish, */
/* distribute, sublicense, and/or sell copies of the Software, and to */
/* permit persons to whom the Software is furnished to do so, subject to */
/* the following conditions: */
/* */
/* The above copyright notice and this permission notice shall be */
/* included in all copies or substantial portions of the Software. */
/* */
/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, */
/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF */
/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. */
/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY */
/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, */
/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE */
/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */
/**************************************************************************/
2018-01-05 07:50:27 +08:00
2017-10-03 05:24:00 +08:00
# include "bindings_generator.h"
2019-09-28 07:07:57 +08:00
# if defined(DEBUG_METHODS_ENABLED) && defined(TOOLS_ENABLED)
2017-10-03 05:24:00 +08:00
2020-11-08 06:33:38 +08:00
# include "core/config/engine.h"
# include "core/core_constants.h"
2018-09-12 00:13:45 +08:00
# include "core/io/compression.h"
2021-06-11 20:51:48 +08:00
# include "core/io/dir_access.h"
# include "core/io/file_access.h"
2018-09-12 00:13:45 +08:00
# include "core/os/os.h"
2021-02-22 18:06:33 +08:00
# include "main/main.h"
2017-10-03 05:24:00 +08:00
# include "../godotsharp_defs.h"
2022-12-07 23:11:39 +08:00
# include "../utils/naming_utils.h"
2017-10-03 05:24:00 +08:00
# include "../utils/path_utils.h"
# include "../utils/string_utils.h"
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
StringBuilder & operator < < ( StringBuilder & r_sb , const String & p_string ) {
r_sb . append ( p_string ) ;
return r_sb ;
}
StringBuilder & operator < < ( StringBuilder & r_sb , const char * p_cstring ) {
r_sb . append ( p_cstring ) ;
return r_sb ;
}
2018-09-04 11:40:41 +08:00
# define CS_INDENT " " // 4 whitespaces
2017-10-03 05:24:00 +08:00
# define INDENT1 CS_INDENT
# define INDENT2 INDENT1 INDENT1
# define INDENT3 INDENT2 INDENT1
# define INDENT4 INDENT3 INDENT1
2022-02-28 04:57:50 +08:00
# define MEMBER_BEGIN "\n" INDENT1
2017-10-03 05:24:00 +08:00
# define OPEN_BLOCK "{\n"
# define CLOSE_BLOCK "}\n"
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
# define OPEN_BLOCK_L1 INDENT1 OPEN_BLOCK
# define OPEN_BLOCK_L2 INDENT2 OPEN_BLOCK
2022-02-28 04:57:50 +08:00
# define CLOSE_BLOCK_L1 INDENT1 CLOSE_BLOCK
2017-10-03 05:24:00 +08:00
# define CLOSE_BLOCK_L2 INDENT2 CLOSE_BLOCK
# define CLOSE_BLOCK_L3 INDENT3 CLOSE_BLOCK
2021-12-29 06:25:16 +08:00
# define BINDINGS_GLOBAL_SCOPE_CLASS "GD"
# define BINDINGS_NATIVE_NAME_FIELD "NativeName"
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
# define CS_PARAM_MEMORYOWN "memoryOwn"
2017-10-03 05:24:00 +08:00
# define CS_PARAM_METHODBIND "method"
# define CS_PARAM_INSTANCE "ptr"
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
# define CS_STATIC_METHOD_GETINSTANCE "GetPtr"
2017-10-03 05:24:00 +08:00
# define CS_METHOD_CALL "Call"
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
# define CS_PROPERTY_SINGLETON "Singleton"
2021-12-29 06:25:16 +08:00
# define CS_METHOD_INVOKE_GODOT_CLASS_METHOD "InvokeGodotClassMethod"
2021-12-29 06:25:16 +08:00
# define CS_METHOD_HAS_GODOT_CLASS_METHOD "HasGodotClassMethod"
2023-01-14 03:59:02 +08:00
# define CS_METHOD_HAS_GODOT_CLASS_SIGNAL "HasGodotClassSignal"
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
# define CS_STATIC_FIELD_NATIVE_CTOR "NativeCtor"
# define CS_STATIC_FIELD_METHOD_BIND_PREFIX "MethodBind"
2021-12-29 06:25:16 +08:00
# define CS_STATIC_FIELD_METHOD_PROXY_NAME_PREFIX "MethodProxyName_"
2023-01-14 03:59:02 +08:00
# define CS_STATIC_FIELD_SIGNAL_PROXY_NAME_PREFIX "SignalProxyName_"
2017-10-03 05:24:00 +08:00
# define ICALL_PREFIX "godot_icall_"
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
# define ICALL_CLASSDB_GET_METHOD "ClassDB_get_method"
# define ICALL_CLASSDB_GET_CONSTRUCTOR "ClassDB_get_constructor"
2018-09-04 11:40:41 +08:00
# define C_LOCAL_RET "ret"
2019-01-18 07:41:41 +08:00
# define C_LOCAL_VARARG_RET "vararg_ret"
2017-10-03 05:24:00 +08:00
# define C_LOCAL_PTRCALL_ARGS "call_args"
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
# define C_CLASS_NATIVE_FUNCS "NativeFuncs"
# define C_NS_MONOUTILS "InteropUtils"
# define C_METHOD_UNMANAGED_GET_MANAGED C_NS_MONOUTILS ".UnmanagedGetManaged"
2021-09-13 02:21:15 +08:00
# define C_METHOD_ENGINE_GET_SINGLETON C_NS_MONOUTILS ".EngineGetSingleton"
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
# define C_NS_MONOMARSHAL "Marshaling"
2021-12-29 06:25:16 +08:00
# define C_METHOD_MONOSTR_TO_GODOT C_NS_MONOMARSHAL ".ConvertStringToNative"
# define C_METHOD_MONOSTR_FROM_GODOT C_NS_MONOMARSHAL ".ConvertStringToManaged"
# define C_METHOD_MONOARRAY_TO(m_type) C_NS_MONOMARSHAL ".ConvertSystemArrayToNative" #m_type
# define C_METHOD_MONOARRAY_FROM(m_type) C_NS_MONOMARSHAL ".ConvertNative" #m_type "ToSystemArray"
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
# define C_METHOD_MANAGED_TO_CALLABLE C_NS_MONOMARSHAL ".ConvertCallableToNative"
# define C_METHOD_MANAGED_FROM_CALLABLE C_NS_MONOMARSHAL ".ConvertCallableToManaged"
# define C_METHOD_MANAGED_TO_SIGNAL C_NS_MONOMARSHAL ".ConvertSignalToNative"
# define C_METHOD_MANAGED_FROM_SIGNAL C_NS_MONOMARSHAL ".ConvertSignalToManaged"
2018-02-22 20:13:51 +08:00
2022-03-19 00:28:48 +08:00
// Types that will be ignored by the generator and won't be available in C#.
2022-09-04 16:32:21 +08:00
const Vector < String > ignored_types = { " PhysicsServer2DExtension " , " PhysicsServer3DExtension " } ;
2022-03-19 00:28:48 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
void BindingsGenerator : : TypeInterface : : postsetup_enum_type ( BindingsGenerator : : TypeInterface & r_enum_itype ) {
// C interface for enums is the same as that of 'uint32_t'. Remember to apply
// any of the changes done here to the 'uint32_t' type interface as well.
r_enum_itype . cs_type = r_enum_itype . proxy_name ;
2021-12-29 06:25:16 +08:00
r_enum_itype . cs_in_expr = " (int)%0 " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
r_enum_itype . cs_out = " %5return (%2)%0(%1); " ;
{
// The expected types for parameters and return value in ptrcall are 'int64_t' or 'uint64_t'.
r_enum_itype . c_in = " %5%0 %1_in = %1; \n " ;
r_enum_itype . c_out = " %5return (%0)%1; \n " ;
r_enum_itype . c_type = " long " ;
r_enum_itype . c_arg_in = " &%s_in " ;
}
r_enum_itype . c_type_in = " int " ;
r_enum_itype . c_type_out = r_enum_itype . c_type_in ;
r_enum_itype . class_doc = & EditorHelp : : get_doc_data ( ) - > class_list [ r_enum_itype . proxy_name ] ;
}
2017-10-03 05:24:00 +08:00
2019-03-19 04:59:35 +08:00
static String fix_doc_description ( const String & p_bbcode ) {
// This seems to be the correct way to do this. It's the same EditorHelp does.
return p_bbcode . dedent ( )
. replace ( " \t " , " " )
. replace ( " \r " , " " )
. strip_edges ( ) ;
}
String BindingsGenerator : : bbcode_to_xml ( const String & p_bbcode , const TypeInterface * p_itype ) {
// Based on the version in EditorHelp
2020-12-15 20:04:21 +08:00
if ( p_bbcode . is_empty ( ) ) {
2019-03-19 04:59:35 +08:00
return String ( ) ;
2020-05-16 10:03:05 +08:00
}
2019-03-19 04:59:35 +08:00
2020-11-29 11:42:06 +08:00
DocTools * doc = EditorHelp : : get_doc_data ( ) ;
2019-03-19 04:59:35 +08:00
String bbcode = p_bbcode ;
StringBuilder xml_output ;
xml_output . append ( " <para> " ) ;
List < String > tag_stack ;
bool code_tag = false ;
2023-01-24 18:40:48 +08:00
bool line_del = false ;
2019-03-19 04:59:35 +08:00
int pos = 0 ;
while ( pos < bbcode . length ( ) ) {
int brk_pos = bbcode . find ( " [ " , pos ) ;
2020-05-16 10:03:05 +08:00
if ( brk_pos < 0 ) {
2019-03-19 04:59:35 +08:00
brk_pos = bbcode . length ( ) ;
2020-05-16 10:03:05 +08:00
}
2019-03-19 04:59:35 +08:00
if ( brk_pos > pos ) {
2023-01-24 18:40:48 +08:00
if ( ! line_del ) {
String text = bbcode . substr ( pos , brk_pos - pos ) ;
if ( code_tag | | tag_stack . size ( ) > 0 ) {
xml_output . append ( text . xml_escape ( ) ) ;
} else {
Vector < String > lines = text . split ( " \n " ) ;
for ( int i = 0 ; i < lines . size ( ) ; i + + ) {
if ( i ! = 0 ) {
xml_output . append ( " <para> " ) ;
}
2019-03-19 04:59:35 +08:00
2023-01-24 18:40:48 +08:00
xml_output . append ( lines [ i ] . xml_escape ( ) ) ;
2019-03-19 04:59:35 +08:00
2023-01-24 18:40:48 +08:00
if ( i ! = lines . size ( ) - 1 ) {
xml_output . append ( " </para> \n " ) ;
}
2020-05-16 10:03:05 +08:00
}
2019-03-19 04:59:35 +08:00
}
}
}
2020-05-16 10:03:05 +08:00
if ( brk_pos = = bbcode . length ( ) ) {
2019-03-19 04:59:35 +08:00
break ; // nothing else to add
2020-05-16 10:03:05 +08:00
}
2019-03-19 04:59:35 +08:00
int brk_end = bbcode . find ( " ] " , brk_pos + 1 ) ;
if ( brk_end = = - 1 ) {
2023-01-24 18:40:48 +08:00
if ( ! line_del ) {
String text = bbcode . substr ( brk_pos , bbcode . length ( ) - brk_pos ) ;
if ( code_tag | | tag_stack . size ( ) > 0 ) {
xml_output . append ( text . xml_escape ( ) ) ;
} else {
Vector < String > lines = text . split ( " \n " ) ;
for ( int i = 0 ; i < lines . size ( ) ; i + + ) {
if ( i ! = 0 ) {
xml_output . append ( " <para> " ) ;
}
2019-03-19 04:59:35 +08:00
2023-01-24 18:40:48 +08:00
xml_output . append ( lines [ i ] . xml_escape ( ) ) ;
2019-03-19 04:59:35 +08:00
2023-01-24 18:40:48 +08:00
if ( i ! = lines . size ( ) - 1 ) {
xml_output . append ( " </para> \n " ) ;
}
2020-05-16 10:03:05 +08:00
}
2019-03-19 04:59:35 +08:00
}
}
break ;
}
String tag = bbcode . substr ( brk_pos + 1 , brk_end - brk_pos - 1 ) ;
if ( tag . begins_with ( " / " ) ) {
bool tag_ok = tag_stack . size ( ) & & tag_stack . front ( ) - > get ( ) = = tag . substr ( 1 , tag . length ( ) ) ;
if ( ! tag_ok ) {
2023-01-24 18:40:48 +08:00
if ( ! line_del ) {
xml_output . append ( " [ " ) ;
}
2019-03-19 04:59:35 +08:00
pos = brk_pos + 1 ;
continue ;
}
tag_stack . pop_front ( ) ;
pos = brk_end + 1 ;
code_tag = false ;
if ( tag = = " /url " ) {
xml_output . append ( " </a> " ) ;
} else if ( tag = = " /code " ) {
xml_output . append ( " </c> " ) ;
} else if ( tag = = " /codeblock " ) {
xml_output . append ( " </code> " ) ;
2023-01-24 18:40:48 +08:00
} else if ( tag = = " /b " ) {
xml_output . append ( " </b> " ) ;
} else if ( tag = = " /i " ) {
xml_output . append ( " </i> " ) ;
} else if ( tag = = " /csharp " ) {
xml_output . append ( " </code> " ) ;
line_del = true ;
} else if ( tag = = " /codeblocks " ) {
line_del = false ;
2019-03-19 04:59:35 +08:00
}
} else if ( code_tag ) {
xml_output . append ( " [ " ) ;
pos = brk_pos + 1 ;
2023-01-24 18:40:48 +08:00
} else if ( tag . begins_with ( " method " ) | | tag . begins_with ( " member " ) | | tag . begins_with ( " signal " ) | | tag . begins_with ( " enum " ) | | tag . begins_with ( " constant " ) | | tag . begins_with ( " theme_item " ) | | tag . begins_with ( " param " ) ) {
2022-02-10 02:01:18 +08:00
const int tag_end = tag . find ( " " ) ;
const String link_tag = tag . substr ( 0 , tag_end ) ;
const String link_target = tag . substr ( tag_end + 1 , tag . length ( ) ) . lstrip ( " " ) ;
2019-03-19 04:59:35 +08:00
2022-03-04 06:35:10 +08:00
const Vector < String > link_target_parts = link_target . split ( " . " ) ;
2019-03-19 04:59:35 +08:00
if ( link_target_parts . size ( ) < = 0 | | link_target_parts . size ( ) > 2 ) {
2019-11-07 00:03:04 +08:00
ERR_PRINT ( " Invalid reference format: ' " + tag + " '. " ) ;
2019-03-19 04:59:35 +08:00
xml_output . append ( " <c> " ) ;
xml_output . append ( tag ) ;
xml_output . append ( " </c> " ) ;
pos = brk_end + 1 ;
continue ;
}
const TypeInterface * target_itype ;
StringName target_cname ;
if ( link_target_parts . size ( ) = = 2 ) {
target_itype = _get_type_or_null ( TypeReference ( link_target_parts [ 0 ] ) ) ;
if ( ! target_itype ) {
target_itype = _get_type_or_null ( TypeReference ( " _ " + link_target_parts [ 0 ] ) ) ;
}
target_cname = link_target_parts [ 1 ] ;
} else {
target_itype = p_itype ;
target_cname = link_target_parts [ 0 ] ;
}
if ( link_tag = = " method " ) {
2022-03-04 06:35:10 +08:00
_append_xml_method ( xml_output , target_itype , target_cname , link_target , link_target_parts ) ;
2019-03-19 04:59:35 +08:00
} else if ( link_tag = = " member " ) {
2022-03-04 06:35:10 +08:00
_append_xml_member ( xml_output , target_itype , target_cname , link_target , link_target_parts ) ;
2019-03-19 04:59:35 +08:00
} else if ( link_tag = = " signal " ) {
2022-03-04 06:35:10 +08:00
_append_xml_signal ( xml_output , target_itype , target_cname , link_target , link_target_parts ) ;
2019-03-19 04:59:35 +08:00
} else if ( link_tag = = " enum " ) {
2022-03-04 06:35:10 +08:00
_append_xml_enum ( xml_output , target_itype , target_cname , link_target , link_target_parts ) ;
2021-08-05 03:17:16 +08:00
} else if ( link_tag = = " constant " ) {
2022-03-04 06:35:10 +08:00
_append_xml_constant ( xml_output , target_itype , target_cname , link_target , link_target_parts ) ;
2022-02-14 23:24:35 +08:00
} else if ( link_tag = = " theme_item " ) {
// We do not declare theme_items in any way in C#, so there is nothing to reference
2022-03-04 06:35:10 +08:00
_append_xml_undeclared ( xml_output , link_target ) ;
2023-01-24 18:40:48 +08:00
} else if ( link_tag = = " param " ) {
_append_xml_undeclared ( xml_output , snake_to_camel_case ( link_target , false ) ) ;
2019-03-19 04:59:35 +08:00
}
pos = brk_end + 1 ;
} else if ( doc - > class_list . has ( tag ) ) {
if ( tag = = " Array " | | tag = = " Dictionary " ) {
xml_output . append ( " <see cref= \" " BINDINGS_NAMESPACE_COLLECTIONS " . " ) ;
xml_output . append ( tag ) ;
xml_output . append ( " \" /> " ) ;
} else if ( tag = = " bool " | | tag = = " int " ) {
xml_output . append ( " <see cref= \" " ) ;
xml_output . append ( tag ) ;
xml_output . append ( " \" /> " ) ;
} else if ( tag = = " float " ) {
xml_output . append ( " <see cref= \" "
# ifdef REAL_T_IS_DOUBLE
" double "
# else
" float "
# endif
" \" /> " ) ;
} else if ( tag = = " Variant " ) {
2023-01-24 18:40:48 +08:00
xml_output . append ( " <see cref= \" Godot.Variant \" /> " ) ;
2019-03-19 04:59:35 +08:00
} else if ( tag = = " String " ) {
xml_output . append ( " <see cref= \" string \" /> " ) ;
} else if ( tag = = " Nil " ) {
xml_output . append ( " <see langword= \" null \" /> " ) ;
} else if ( tag . begins_with ( " @ " ) ) {
2019-03-28 03:01:16 +08:00
// @GlobalScope, @GDScript, etc
2019-03-19 04:59:35 +08:00
xml_output . append ( " <c> " ) ;
xml_output . append ( tag ) ;
xml_output . append ( " </c> " ) ;
2020-02-18 05:06:54 +08:00
} else if ( tag = = " PackedByteArray " ) {
2021-12-29 06:25:16 +08:00
xml_output . append ( " <see cref= \" byte \" />[] " ) ;
Variant: Added 64-bit packed arrays, renamed Variant::REAL to FLOAT.
- Renames PackedIntArray to PackedInt32Array.
- Renames PackedFloatArray to PackedFloat32Array.
- Adds PackedInt64Array and PackedFloat64Array.
- Renames Variant::REAL to Variant::FLOAT for consistency.
Packed arrays are for storing large amount of data and creating stuff like
meshes, buffers. textures, etc. Forcing them to be 64 is a huge waste of
memory. That said, many users requested the ability to have 64 bits packed
arrays for their games, so this is just an optional added type.
For Variant, the float datatype is always 64 bits, and exposed as `float`.
We still have `real_t` which is the datatype that can change from 32 to 64
bits depending on a compile flag (not entirely working right now, but that's
the idea). It affects math related datatypes and code only.
Neither Variant nor PackedArray make use of real_t, which is only intended
for math precision, so the term is removed from there to keep only float.
2020-02-25 02:20:53 +08:00
} else if ( tag = = " PackedInt32Array " ) {
2021-12-29 06:25:16 +08:00
xml_output . append ( " <see cref= \" int \" />[] " ) ;
2020-03-15 02:20:17 +08:00
} else if ( tag = = " PackedInt64Array " ) {
2021-12-29 06:25:16 +08:00
xml_output . append ( " <see cref= \" long \" />[] " ) ;
Variant: Added 64-bit packed arrays, renamed Variant::REAL to FLOAT.
- Renames PackedIntArray to PackedInt32Array.
- Renames PackedFloatArray to PackedFloat32Array.
- Adds PackedInt64Array and PackedFloat64Array.
- Renames Variant::REAL to Variant::FLOAT for consistency.
Packed arrays are for storing large amount of data and creating stuff like
meshes, buffers. textures, etc. Forcing them to be 64 is a huge waste of
memory. That said, many users requested the ability to have 64 bits packed
arrays for their games, so this is just an optional added type.
For Variant, the float datatype is always 64 bits, and exposed as `float`.
We still have `real_t` which is the datatype that can change from 32 to 64
bits depending on a compile flag (not entirely working right now, but that's
the idea). It affects math related datatypes and code only.
Neither Variant nor PackedArray make use of real_t, which is only intended
for math precision, so the term is removed from there to keep only float.
2020-02-25 02:20:53 +08:00
} else if ( tag = = " PackedFloat32Array " ) {
2021-12-29 06:25:16 +08:00
xml_output . append ( " <see cref= \" float \" />[] " ) ;
2020-03-15 02:20:17 +08:00
} else if ( tag = = " PackedFloat64Array " ) {
2021-12-29 06:25:16 +08:00
xml_output . append ( " <see cref= \" double \" />[] " ) ;
2020-02-18 05:06:54 +08:00
} else if ( tag = = " PackedStringArray " ) {
2021-12-29 06:25:16 +08:00
xml_output . append ( " <see cref= \" string \" />[] " ) ;
2020-02-18 05:06:54 +08:00
} else if ( tag = = " PackedVector2Array " ) {
2021-12-29 06:25:16 +08:00
xml_output . append ( " <see cref= \" " BINDINGS_NAMESPACE " .Vector2 \" />[] " ) ;
2020-02-18 05:06:54 +08:00
} else if ( tag = = " PackedVector3Array " ) {
2021-12-29 06:25:16 +08:00
xml_output . append ( " <see cref= \" " BINDINGS_NAMESPACE " .Vector3 \" />[] " ) ;
2020-02-18 05:06:54 +08:00
} else if ( tag = = " PackedColorArray " ) {
2021-12-29 06:25:16 +08:00
xml_output . append ( " <see cref= \" " BINDINGS_NAMESPACE " .Color \" />[] " ) ;
2019-03-19 04:59:35 +08:00
} else {
const TypeInterface * target_itype = _get_type_or_null ( TypeReference ( tag ) ) ;
if ( ! target_itype ) {
target_itype = _get_type_or_null ( TypeReference ( " _ " + tag ) ) ;
}
if ( target_itype ) {
xml_output . append ( " <see cref= \" " BINDINGS_NAMESPACE " . " ) ;
xml_output . append ( target_itype - > proxy_name ) ;
xml_output . append ( " \" /> " ) ;
} else {
2019-11-07 00:03:04 +08:00
ERR_PRINT ( " Cannot resolve type reference in documentation: ' " + tag + " '. " ) ;
2019-03-19 04:59:35 +08:00
xml_output . append ( " <c> " ) ;
xml_output . append ( tag ) ;
xml_output . append ( " </c> " ) ;
}
}
pos = brk_end + 1 ;
} else if ( tag = = " b " ) {
2023-01-24 18:40:48 +08:00
xml_output . append ( " <b> " ) ;
2019-03-19 04:59:35 +08:00
pos = brk_end + 1 ;
tag_stack . push_front ( tag ) ;
} else if ( tag = = " i " ) {
2023-01-24 18:40:48 +08:00
xml_output . append ( " <i> " ) ;
2019-03-19 04:59:35 +08:00
pos = brk_end + 1 ;
tag_stack . push_front ( tag ) ;
} else if ( tag = = " code " ) {
xml_output . append ( " <c> " ) ;
code_tag = true ;
pos = brk_end + 1 ;
tag_stack . push_front ( tag ) ;
} else if ( tag = = " codeblock " ) {
xml_output . append ( " <code> " ) ;
2023-01-24 18:40:48 +08:00
code_tag = true ;
pos = brk_end + 1 ;
tag_stack . push_front ( tag ) ;
} else if ( tag = = " codeblocks " ) {
line_del = true ;
pos = brk_end + 1 ;
tag_stack . push_front ( tag ) ;
} else if ( tag = = " csharp " ) {
xml_output . append ( " <code> " ) ;
line_del = false ;
2019-03-19 04:59:35 +08:00
code_tag = true ;
pos = brk_end + 1 ;
tag_stack . push_front ( tag ) ;
2020-03-10 18:41:36 +08:00
} else if ( tag = = " kbd " ) {
// keyboard combinations are not supported in xml comments
pos = brk_end + 1 ;
tag_stack . push_front ( tag ) ;
2019-03-19 04:59:35 +08:00
} else if ( tag = = " center " ) {
2020-03-10 18:41:36 +08:00
// center alignment is not supported in xml comments
2019-03-19 04:59:35 +08:00
pos = brk_end + 1 ;
tag_stack . push_front ( tag ) ;
} else if ( tag = = " br " ) {
xml_output . append ( " \n " ) ; // FIXME: Should use <para> instead. Luckily this tag isn't used for now.
pos = brk_end + 1 ;
} else if ( tag = = " u " ) {
2023-01-24 18:40:48 +08:00
// underline is not supported in Rider xml comments
2019-03-19 04:59:35 +08:00
pos = brk_end + 1 ;
tag_stack . push_front ( tag ) ;
} else if ( tag = = " s " ) {
// strikethrough is not supported in xml comments
pos = brk_end + 1 ;
tag_stack . push_front ( tag ) ;
} else if ( tag = = " url " ) {
int end = bbcode . find ( " [ " , brk_end ) ;
2020-05-16 10:03:05 +08:00
if ( end = = - 1 ) {
2019-03-19 04:59:35 +08:00
end = bbcode . length ( ) ;
2020-05-16 10:03:05 +08:00
}
2019-03-19 04:59:35 +08:00
String url = bbcode . substr ( brk_end + 1 , end - brk_end - 1 ) ;
xml_output . append ( " <a href= \" " ) ;
xml_output . append ( url ) ;
xml_output . append ( " \" > " ) ;
xml_output . append ( url ) ;
pos = brk_end + 1 ;
tag_stack . push_front ( tag ) ;
} else if ( tag . begins_with ( " url= " ) ) {
String url = tag . substr ( 4 , tag . length ( ) ) ;
xml_output . append ( " <a href= \" " ) ;
xml_output . append ( url ) ;
xml_output . append ( " \" > " ) ;
pos = brk_end + 1 ;
tag_stack . push_front ( " url " ) ;
} else if ( tag = = " img " ) {
int end = bbcode . find ( " [ " , brk_end ) ;
2020-05-16 10:03:05 +08:00
if ( end = = - 1 ) {
2019-03-19 04:59:35 +08:00
end = bbcode . length ( ) ;
2020-05-16 10:03:05 +08:00
}
2019-03-19 04:59:35 +08:00
String image = bbcode . substr ( brk_end + 1 , end - brk_end - 1 ) ;
// Not supported. Just append the bbcode.
xml_output . append ( " [img] " ) ;
xml_output . append ( image ) ;
xml_output . append ( " [/img] " ) ;
pos = end ;
tag_stack . push_front ( tag ) ;
} else if ( tag . begins_with ( " color= " ) ) {
// Not supported.
pos = brk_end + 1 ;
tag_stack . push_front ( " color " ) ;
} else if ( tag . begins_with ( " font= " ) ) {
// Not supported.
pos = brk_end + 1 ;
tag_stack . push_front ( " font " ) ;
} else {
2023-01-24 18:40:48 +08:00
if ( ! line_del ) {
xml_output . append ( " [ " ) ; // ignore
}
2019-03-19 04:59:35 +08:00
pos = brk_pos + 1 ;
}
}
xml_output . append ( " </para> " ) ;
return xml_output . as_string ( ) ;
}
2022-03-04 06:35:10 +08:00
void BindingsGenerator : : _append_xml_method ( StringBuilder & p_xml_output , const TypeInterface * p_target_itype , const StringName & p_target_cname , const String & p_link_target , const Vector < String > & p_link_target_parts ) {
if ( p_link_target_parts [ 0 ] = = name_cache . type_at_GlobalScope ) {
if ( OS : : get_singleton ( ) - > is_stdout_verbose ( ) ) {
OS : : get_singleton ( ) - > print ( " Cannot resolve @GlobalScope method reference in documentation: %s \n " , p_link_target . utf8 ( ) . get_data ( ) ) ;
}
// TODO Map what we can
_append_xml_undeclared ( p_xml_output , p_link_target ) ;
} else if ( ! p_target_itype | | ! p_target_itype - > is_object_type ) {
if ( OS : : get_singleton ( ) - > is_stdout_verbose ( ) ) {
if ( p_target_itype ) {
2022-12-07 23:16:51 +08:00
OS : : get_singleton ( ) - > print ( " Cannot resolve method reference for non-GodotObject type in documentation: %s \n " , p_link_target . utf8 ( ) . get_data ( ) ) ;
2022-03-04 06:35:10 +08:00
} else {
OS : : get_singleton ( ) - > print ( " Cannot resolve type from method reference in documentation: %s \n " , p_link_target . utf8 ( ) . get_data ( ) ) ;
}
}
// TODO Map what we can
_append_xml_undeclared ( p_xml_output , p_link_target ) ;
} else {
if ( p_target_cname = = " _init " ) {
// The _init method is not declared in C#, reference the constructor instead
p_xml_output . append ( " <see cref= \" " BINDINGS_NAMESPACE " . " ) ;
p_xml_output . append ( p_target_itype - > proxy_name ) ;
p_xml_output . append ( " . " ) ;
p_xml_output . append ( p_target_itype - > proxy_name ) ;
p_xml_output . append ( " () \" /> " ) ;
} else {
const MethodInterface * target_imethod = p_target_itype - > find_method_by_name ( p_target_cname ) ;
if ( target_imethod ) {
p_xml_output . append ( " <see cref= \" " BINDINGS_NAMESPACE " . " ) ;
p_xml_output . append ( p_target_itype - > proxy_name ) ;
p_xml_output . append ( " . " ) ;
p_xml_output . append ( target_imethod - > proxy_name ) ;
p_xml_output . append ( " \" /> " ) ;
} else {
2023-03-24 18:40:54 +08:00
if ( ! p_target_itype - > is_intentionally_ignored ( p_link_target ) ) {
ERR_PRINT ( " Cannot resolve method reference in documentation: ' " + p_link_target + " '. " ) ;
}
2022-03-04 06:35:10 +08:00
_append_xml_undeclared ( p_xml_output , p_link_target ) ;
}
}
}
}
void BindingsGenerator : : _append_xml_member ( StringBuilder & p_xml_output , const TypeInterface * p_target_itype , const StringName & p_target_cname , const String & p_link_target , const Vector < String > & p_link_target_parts ) {
if ( p_link_target . find ( " / " ) > = 0 ) {
// Properties with '/' (slash) in the name are not declared in C#, so there is nothing to reference.
_append_xml_undeclared ( p_xml_output , p_link_target ) ;
} else if ( ! p_target_itype | | ! p_target_itype - > is_object_type ) {
if ( OS : : get_singleton ( ) - > is_stdout_verbose ( ) ) {
if ( p_target_itype ) {
2022-12-07 23:16:51 +08:00
OS : : get_singleton ( ) - > print ( " Cannot resolve member reference for non-GodotObject type in documentation: %s \n " , p_link_target . utf8 ( ) . get_data ( ) ) ;
2022-03-04 06:35:10 +08:00
} else {
OS : : get_singleton ( ) - > print ( " Cannot resolve type from member reference in documentation: %s \n " , p_link_target . utf8 ( ) . get_data ( ) ) ;
}
}
// TODO Map what we can
_append_xml_undeclared ( p_xml_output , p_link_target ) ;
} else {
const TypeInterface * current_itype = p_target_itype ;
const PropertyInterface * target_iprop = nullptr ;
while ( target_iprop = = nullptr & & current_itype ! = nullptr ) {
target_iprop = current_itype - > find_property_by_name ( p_target_cname ) ;
if ( target_iprop = = nullptr ) {
current_itype = _get_type_or_null ( TypeReference ( current_itype - > base_name ) ) ;
}
}
if ( target_iprop ) {
p_xml_output . append ( " <see cref= \" " BINDINGS_NAMESPACE " . " ) ;
p_xml_output . append ( current_itype - > proxy_name ) ;
p_xml_output . append ( " . " ) ;
p_xml_output . append ( target_iprop - > proxy_name ) ;
p_xml_output . append ( " \" /> " ) ;
} else {
2023-03-24 18:40:54 +08:00
if ( ! p_target_itype - > is_intentionally_ignored ( p_link_target ) ) {
ERR_PRINT ( " Cannot resolve member reference in documentation: ' " + p_link_target + " '. " ) ;
}
2022-03-04 06:35:10 +08:00
_append_xml_undeclared ( p_xml_output , p_link_target ) ;
}
}
}
void BindingsGenerator : : _append_xml_signal ( StringBuilder & p_xml_output , const TypeInterface * p_target_itype , const StringName & p_target_cname , const String & p_link_target , const Vector < String > & p_link_target_parts ) {
if ( ! p_target_itype | | ! p_target_itype - > is_object_type ) {
if ( OS : : get_singleton ( ) - > is_stdout_verbose ( ) ) {
if ( p_target_itype ) {
2022-12-07 23:16:51 +08:00
OS : : get_singleton ( ) - > print ( " Cannot resolve signal reference for non-GodotObject type in documentation: %s \n " , p_link_target . utf8 ( ) . get_data ( ) ) ;
2022-03-04 06:35:10 +08:00
} else {
OS : : get_singleton ( ) - > print ( " Cannot resolve type from signal reference in documentation: %s \n " , p_link_target . utf8 ( ) . get_data ( ) ) ;
}
}
// TODO Map what we can
_append_xml_undeclared ( p_xml_output , p_link_target ) ;
} else {
const SignalInterface * target_isignal = p_target_itype - > find_signal_by_name ( p_target_cname ) ;
if ( target_isignal ) {
p_xml_output . append ( " <see cref= \" " BINDINGS_NAMESPACE " . " ) ;
p_xml_output . append ( p_target_itype - > proxy_name ) ;
p_xml_output . append ( " . " ) ;
p_xml_output . append ( target_isignal - > proxy_name ) ;
p_xml_output . append ( " \" /> " ) ;
} else {
2023-03-24 18:40:54 +08:00
if ( ! p_target_itype - > is_intentionally_ignored ( p_link_target ) ) {
ERR_PRINT ( " Cannot resolve signal reference in documentation: ' " + p_link_target + " '. " ) ;
}
2022-03-04 06:35:10 +08:00
_append_xml_undeclared ( p_xml_output , p_link_target ) ;
}
}
}
void BindingsGenerator : : _append_xml_enum ( StringBuilder & p_xml_output , const TypeInterface * p_target_itype , const StringName & p_target_cname , const String & p_link_target , const Vector < String > & p_link_target_parts ) {
const StringName search_cname = ! p_target_itype ? p_target_cname : StringName ( p_target_itype - > name + " . " + ( String ) p_target_cname ) ;
2022-05-13 21:04:37 +08:00
HashMap < StringName , TypeInterface > : : ConstIterator enum_match = enum_types . find ( search_cname ) ;
2022-03-04 06:35:10 +08:00
if ( ! enum_match & & search_cname ! = p_target_cname ) {
enum_match = enum_types . find ( p_target_cname ) ;
}
if ( enum_match ) {
2022-05-13 21:04:37 +08:00
const TypeInterface & target_enum_itype = enum_match - > value ;
2022-03-04 06:35:10 +08:00
p_xml_output . append ( " <see cref= \" " BINDINGS_NAMESPACE " . " ) ;
p_xml_output . append ( target_enum_itype . proxy_name ) ; // Includes nesting class if any
p_xml_output . append ( " \" /> " ) ;
} else {
2023-03-24 18:40:54 +08:00
if ( ! p_target_itype - > is_intentionally_ignored ( p_link_target ) ) {
ERR_PRINT ( " Cannot resolve enum reference in documentation: ' " + p_link_target + " '. " ) ;
}
2022-03-04 06:35:10 +08:00
_append_xml_undeclared ( p_xml_output , p_link_target ) ;
}
}
void BindingsGenerator : : _append_xml_constant ( StringBuilder & p_xml_output , const TypeInterface * p_target_itype , const StringName & p_target_cname , const String & p_link_target , const Vector < String > & p_link_target_parts ) {
if ( p_link_target_parts [ 0 ] = = name_cache . type_at_GlobalScope ) {
_append_xml_constant_in_global_scope ( p_xml_output , p_target_cname , p_link_target ) ;
} else if ( ! p_target_itype | | ! p_target_itype - > is_object_type ) {
// Search in @GlobalScope as a last resort if no class was specified
if ( p_link_target_parts . size ( ) = = 1 ) {
_append_xml_constant_in_global_scope ( p_xml_output , p_target_cname , p_link_target ) ;
return ;
}
if ( OS : : get_singleton ( ) - > is_stdout_verbose ( ) ) {
if ( p_target_itype ) {
2022-12-07 23:16:51 +08:00
OS : : get_singleton ( ) - > print ( " Cannot resolve constant reference for non-GodotObject type in documentation: %s \n " , p_link_target . utf8 ( ) . get_data ( ) ) ;
2022-03-04 06:35:10 +08:00
} else {
OS : : get_singleton ( ) - > print ( " Cannot resolve type from constant reference in documentation: %s \n " , p_link_target . utf8 ( ) . get_data ( ) ) ;
}
}
// TODO Map what we can
_append_xml_undeclared ( p_xml_output , p_link_target ) ;
} else {
// Try to find the constant in the current class
const ConstantInterface * target_iconst = find_constant_by_name ( p_target_cname , p_target_itype - > constants ) ;
if ( target_iconst ) {
// Found constant in current class
p_xml_output . append ( " <see cref= \" " BINDINGS_NAMESPACE " . " ) ;
p_xml_output . append ( p_target_itype - > proxy_name ) ;
p_xml_output . append ( " . " ) ;
p_xml_output . append ( target_iconst - > proxy_name ) ;
p_xml_output . append ( " \" /> " ) ;
} else {
// Try to find as enum constant in the current class
const EnumInterface * target_ienum = nullptr ;
for ( const EnumInterface & ienum : p_target_itype - > enums ) {
target_ienum = & ienum ;
target_iconst = find_constant_by_name ( p_target_cname , target_ienum - > constants ) ;
if ( target_iconst ) {
break ;
}
}
if ( target_iconst ) {
p_xml_output . append ( " <see cref= \" " BINDINGS_NAMESPACE " . " ) ;
p_xml_output . append ( p_target_itype - > proxy_name ) ;
p_xml_output . append ( " . " ) ;
p_xml_output . append ( target_ienum - > cname ) ;
p_xml_output . append ( " . " ) ;
p_xml_output . append ( target_iconst - > proxy_name ) ;
p_xml_output . append ( " \" /> " ) ;
} else if ( p_link_target_parts . size ( ) = = 1 ) {
// Also search in @GlobalScope as a last resort if no class was specified
_append_xml_constant_in_global_scope ( p_xml_output , p_target_cname , p_link_target ) ;
} else {
2023-03-24 18:40:54 +08:00
if ( ! p_target_itype - > is_intentionally_ignored ( p_link_target ) ) {
ERR_PRINT ( " Cannot resolve constant reference in documentation: ' " + p_link_target + " '. " ) ;
}
2022-03-04 06:35:10 +08:00
_append_xml_undeclared ( p_xml_output , p_link_target ) ;
}
}
}
}
void BindingsGenerator : : _append_xml_constant_in_global_scope ( StringBuilder & p_xml_output , const String & p_target_cname , const String & p_link_target ) {
// Try to find as a global constant
const ConstantInterface * target_iconst = find_constant_by_name ( p_target_cname , global_constants ) ;
if ( target_iconst ) {
// Found global constant
p_xml_output . append ( " <see cref= \" " BINDINGS_NAMESPACE " . " BINDINGS_GLOBAL_SCOPE_CLASS " . " ) ;
p_xml_output . append ( target_iconst - > proxy_name ) ;
p_xml_output . append ( " \" /> " ) ;
} else {
// Try to find as global enum constant
const EnumInterface * target_ienum = nullptr ;
for ( const EnumInterface & ienum : global_enums ) {
target_ienum = & ienum ;
target_iconst = find_constant_by_name ( p_target_cname , target_ienum - > constants ) ;
if ( target_iconst ) {
break ;
}
}
if ( target_iconst ) {
p_xml_output . append ( " <see cref= \" " BINDINGS_NAMESPACE " . " ) ;
p_xml_output . append ( target_ienum - > cname ) ;
p_xml_output . append ( " . " ) ;
p_xml_output . append ( target_iconst - > proxy_name ) ;
p_xml_output . append ( " \" /> " ) ;
} else {
ERR_PRINT ( " Cannot resolve global constant reference in documentation: ' " + p_link_target + " '. " ) ;
_append_xml_undeclared ( p_xml_output , p_link_target ) ;
}
}
}
void BindingsGenerator : : _append_xml_undeclared ( StringBuilder & p_xml_output , const String & p_link_target ) {
p_xml_output . append ( " <c> " ) ;
p_xml_output . append ( p_link_target ) ;
p_xml_output . append ( " </c> " ) ;
}
2018-10-19 01:41:10 +08:00
int BindingsGenerator : : _determine_enum_prefix ( const EnumInterface & p_ienum ) {
2020-12-15 20:04:21 +08:00
CRASH_COND ( p_ienum . constants . is_empty ( ) ) ;
2017-12-24 10:17:48 +08:00
2018-10-19 01:41:10 +08:00
const ConstantInterface & front_iconstant = p_ienum . constants . front ( ) - > get ( ) ;
Vector < String > front_parts = front_iconstant . name . split ( " _ " , /* p_allow_empty: */ true ) ;
int candidate_len = front_parts . size ( ) - 1 ;
2017-12-24 10:17:48 +08:00
2020-05-16 10:03:05 +08:00
if ( candidate_len = = 0 ) {
2018-10-19 01:41:10 +08:00
return 0 ;
2020-05-16 10:03:05 +08:00
}
2018-10-19 01:41:10 +08:00
2021-07-27 03:31:17 +08:00
for ( const ConstantInterface & iconstant : p_ienum . constants ) {
2018-10-19 01:41:10 +08:00
Vector < String > parts = iconstant . name . split ( " _ " , /* p_allow_empty: */ true ) ;
int i ;
for ( i = 0 ; i < candidate_len & & i < parts . size ( ) ; i + + ) {
if ( front_parts [ i ] ! = parts [ i ] ) {
// HARDCODED: Some Flag enums have the prefix 'FLAG_' for everything except 'FLAGS_DEFAULT' (same for 'METHOD_FLAG_' and'METHOD_FLAGS_DEFAULT').
bool hardcoded_exc = ( i = = candidate_len - 1 & & ( ( front_parts [ i ] = = " FLAGS " & & parts [ i ] = = " FLAG " ) | | ( front_parts [ i ] = = " FLAG " & & parts [ i ] = = " FLAGS " ) ) ) ;
2020-05-16 10:03:05 +08:00
if ( ! hardcoded_exc ) {
2018-10-19 01:41:10 +08:00
break ;
2020-05-16 10:03:05 +08:00
}
2018-10-19 01:41:10 +08:00
}
2017-12-24 10:17:48 +08:00
}
2018-10-19 01:41:10 +08:00
candidate_len = i ;
2020-05-16 10:03:05 +08:00
if ( candidate_len = = 0 ) {
2018-10-19 01:41:10 +08:00
return 0 ;
2020-05-16 10:03:05 +08:00
}
2017-12-24 10:17:48 +08:00
}
2018-10-19 01:41:10 +08:00
return candidate_len ;
}
void BindingsGenerator : : _apply_prefix_to_enum_constants ( BindingsGenerator : : EnumInterface & p_ienum , int p_prefix_length ) {
if ( p_prefix_length > 0 ) {
2021-07-26 00:45:00 +08:00
for ( ConstantInterface & iconstant : p_ienum . constants ) {
2018-10-19 01:41:10 +08:00
int curr_prefix_length = p_prefix_length ;
2021-07-26 00:45:00 +08:00
String constant_name = iconstant . name ;
2018-10-19 01:41:10 +08:00
Vector < String > parts = constant_name . split ( " _ " , /* p_allow_empty: */ true ) ;
2020-05-16 10:03:05 +08:00
if ( parts . size ( ) < = curr_prefix_length ) {
2018-10-19 01:41:10 +08:00
continue ;
2020-05-16 10:03:05 +08:00
}
2018-10-19 01:41:10 +08:00
2022-02-04 16:32:20 +08:00
if ( is_digit ( parts [ curr_prefix_length ] [ 0 ] ) ) {
2018-10-19 01:41:10 +08:00
// The name of enum constants may begin with a numeric digit when strip from the enum prefix,
// so we make the prefix for this constant one word shorter in those cases.
for ( curr_prefix_length = curr_prefix_length - 1 ; curr_prefix_length > 0 ; curr_prefix_length - - ) {
2022-02-04 16:32:20 +08:00
if ( ! is_digit ( parts [ curr_prefix_length ] [ 0 ] ) ) {
2018-10-19 01:41:10 +08:00
break ;
2020-05-16 10:03:05 +08:00
}
2018-10-19 01:41:10 +08:00
}
}
constant_name = " " ;
for ( int i = curr_prefix_length ; i < parts . size ( ) ; i + + ) {
2020-05-16 10:03:05 +08:00
if ( i > curr_prefix_length ) {
2018-10-19 01:41:10 +08:00
constant_name + = " _ " ;
2020-05-16 10:03:05 +08:00
}
2018-10-19 01:41:10 +08:00
constant_name + = parts [ i ] ;
}
2021-07-26 00:45:00 +08:00
iconstant . proxy_name = snake_to_pascal_case ( constant_name , true ) ;
2018-10-19 01:41:10 +08:00
}
}
2017-12-24 10:17:48 +08:00
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
Error BindingsGenerator : : _populate_method_icalls_table ( const TypeInterface & p_itype ) {
2021-07-16 11:45:57 +08:00
for ( const MethodInterface & imethod : p_itype . methods ) {
2020-05-16 10:03:05 +08:00
if ( imethod . is_virtual ) {
2017-10-03 05:24:00 +08:00
continue ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
const TypeInterface * return_type = _get_type_or_null ( imethod . return_type ) ;
ERR_FAIL_NULL_V ( return_type , ERR_BUG ) ; // Return type not found
2017-10-03 05:24:00 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
String im_unique_sig = get_ret_unique_sig ( return_type ) + " ,CallMethodBind " ;
2018-01-26 06:44:37 +08:00
2022-03-21 08:25:50 +08:00
if ( ! imethod . is_static ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
im_unique_sig + = " ,CallInstance " ;
2022-03-21 08:25:50 +08:00
}
2017-10-03 05:24:00 +08:00
// Get arguments information
2021-07-26 00:45:00 +08:00
for ( const ArgumentInterface & iarg : imethod . arguments ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
const TypeInterface * arg_type = _get_type_or_null ( iarg . type ) ;
ERR_FAIL_NULL_V ( arg_type , ERR_BUG ) ; // Argument type not found
2017-10-03 05:24:00 +08:00
2018-09-04 11:40:41 +08:00
im_unique_sig + = " , " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
im_unique_sig + = get_arg_unique_sig ( * arg_type ) ;
2019-11-15 04:48:49 +08:00
}
2018-09-04 11:40:41 +08:00
// godot_icall_{argc}_{icallcount}
2018-01-26 06:44:37 +08:00
String icall_method = ICALL_PREFIX ;
2018-09-04 11:40:41 +08:00
icall_method + = itos ( imethod . arguments . size ( ) ) ;
icall_method + = " _ " ;
icall_method + = itos ( method_icalls . size ( ) ) ;
2017-10-03 05:24:00 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
InternalCall im_icall = InternalCall ( p_itype . api_type , icall_method , im_unique_sig ) ;
im_icall . is_vararg = imethod . is_vararg ;
im_icall . is_static = imethod . is_static ;
im_icall . return_type = imethod . return_type ;
for ( const List < ArgumentInterface > : : Element * F = imethod . arguments . front ( ) ; F ; F = F - > next ( ) ) {
im_icall . argument_types . push_back ( F - > get ( ) . type ) ;
}
2017-10-03 05:24:00 +08:00
2018-09-04 11:40:41 +08:00
List < InternalCall > : : Element * match = method_icalls . find ( im_icall ) ;
2017-10-03 05:24:00 +08:00
2018-09-04 11:40:41 +08:00
if ( match ) {
2020-05-16 10:03:05 +08:00
if ( p_itype . api_type ! = ClassDB : : API_EDITOR ) {
2018-09-04 11:40:41 +08:00
match - > get ( ) . editor_only = false ;
2020-05-16 10:03:05 +08:00
}
2021-07-26 00:45:00 +08:00
method_icalls_map . insert ( & imethod , & match - > get ( ) ) ;
2017-10-03 05:24:00 +08:00
} else {
2018-09-04 11:40:41 +08:00
List < InternalCall > : : Element * added = method_icalls . push_back ( im_icall ) ;
2021-07-26 00:45:00 +08:00
method_icalls_map . insert ( & imethod , & added - > get ( ) ) ;
2017-10-03 05:24:00 +08:00
}
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
return OK ;
2017-10-03 05:24:00 +08:00
}
2021-01-13 18:17:39 +08:00
void BindingsGenerator : : _generate_array_extensions ( StringBuilder & p_output ) {
2022-02-28 04:57:50 +08:00
p_output . append ( " namespace " BINDINGS_NAMESPACE " ; \n \n " ) ;
2021-01-13 18:17:39 +08:00
p_output . append ( " using System; \n \n " ) ;
// The class where we put the extensions doesn't matter, so just use "GD".
2022-02-28 04:57:50 +08:00
p_output . append ( " public static partial class " BINDINGS_GLOBAL_SCOPE_CLASS " \n { " ) ;
2021-01-13 18:17:39 +08:00
# define ARRAY_IS_EMPTY(m_type) \
2022-02-28 04:57:50 +08:00
p_output . append ( " \n " INDENT1 " /// <summary> \n " ) ; \
p_output . append ( INDENT1 " /// Returns true if this " # m_type " array is empty or doesn't exist. \n " ) ; \
p_output . append ( INDENT1 " /// </summary> \n " ) ; \
p_output . append ( INDENT1 " /// <param name= \" instance \" >The " # m_type " array check.</param> \n " ) ; \
p_output . append ( INDENT1 " /// <returns>Whether or not the array is empty.</returns> \n " ) ; \
p_output . append ( INDENT1 " public static bool IsEmpty(this " # m_type " [] instance) \n " ) ; \
p_output . append ( OPEN_BLOCK_L1 ) ; \
p_output . append ( INDENT2 " return instance == null || instance.Length == 0; \n " ) ; \
p_output . append ( INDENT1 CLOSE_BLOCK ) ;
2021-01-13 18:17:39 +08:00
# define ARRAY_JOIN(m_type) \
2022-02-28 04:57:50 +08:00
p_output . append ( " \n " INDENT1 " /// <summary> \n " ) ; \
p_output . append ( INDENT1 " /// Converts this " # m_type " array to a string delimited by the given string. \n " ) ; \
p_output . append ( INDENT1 " /// </summary> \n " ) ; \
p_output . append ( INDENT1 " /// <param name= \" instance \" >The " # m_type " array to convert.</param> \n " ) ; \
p_output . append ( INDENT1 " /// <param name= \" delimiter \" >The delimiter to use between items.</param> \n " ) ; \
p_output . append ( INDENT1 " /// <returns>A single string with all items.</returns> \n " ) ; \
p_output . append ( INDENT1 " public static string Join(this " # m_type " [] instance, string delimiter = \" , \" ) \n " ) ; \
p_output . append ( OPEN_BLOCK_L1 ) ; \
p_output . append ( INDENT2 " return String.Join(delimiter, instance); \n " ) ; \
p_output . append ( INDENT1 CLOSE_BLOCK ) ;
2021-01-13 18:17:39 +08:00
# define ARRAY_STRINGIFY(m_type) \
2022-02-28 04:57:50 +08:00
p_output . append ( " \n " INDENT1 " /// <summary> \n " ) ; \
p_output . append ( INDENT1 " /// Converts this " # m_type " array to a string with brackets. \n " ) ; \
p_output . append ( INDENT1 " /// </summary> \n " ) ; \
p_output . append ( INDENT1 " /// <param name= \" instance \" >The " # m_type " array to convert.</param> \n " ) ; \
p_output . append ( INDENT1 " /// <returns>A single string with all items.</returns> \n " ) ; \
p_output . append ( INDENT1 " public static string Stringify(this " # m_type " [] instance) \n " ) ; \
p_output . append ( OPEN_BLOCK_L1 ) ; \
p_output . append ( INDENT2 " return \" [ \" + instance.Join() + \" ] \" ; \n " ) ; \
p_output . append ( INDENT1 CLOSE_BLOCK ) ;
2021-01-13 18:17:39 +08:00
# define ARRAY_ALL(m_type) \
ARRAY_IS_EMPTY ( m_type ) \
ARRAY_JOIN ( m_type ) \
ARRAY_STRINGIFY ( m_type )
ARRAY_ALL ( byte ) ;
ARRAY_ALL ( int ) ;
ARRAY_ALL ( long ) ;
ARRAY_ALL ( float ) ;
ARRAY_ALL ( double ) ;
ARRAY_ALL ( string ) ;
ARRAY_ALL ( Color ) ;
ARRAY_ALL ( Vector2 ) ;
2022-12-07 23:11:39 +08:00
ARRAY_ALL ( Vector2I ) ;
2021-01-13 18:17:39 +08:00
ARRAY_ALL ( Vector3 ) ;
2022-12-07 23:11:39 +08:00
ARRAY_ALL ( Vector3I ) ;
Implement Vector4, Vector4i, Projection
Implement built-in classes Vector4, Vector4i and Projection.
* Two versions of Vector4 (float and integer).
* A Projection class, which is a 4x4 matrix specialized in projection types.
These types have been requested for a long time, but given they were very corner case they were not added before.
Because in Godot 4, reimplementing parts of the rendering engine is now possible, access to these types (heavily used by the rendering code) becomes a necessity.
**Q**: Why Projection and not Matrix4?
**A**: Godot does not use Matrix2, Matrix3, Matrix4x3, etc. naming convention because, within the engine, these types always have a *purpose*. As such, Godot names them: Transform2D, Transform3D or Basis. In this case, this 4x4 matrix is _always_ used as a _Projection_, hence the naming.
2022-07-20 07:11:13 +08:00
ARRAY_ALL ( Vector4 ) ;
2022-12-07 23:11:39 +08:00
ARRAY_ALL ( Vector4I ) ;
2021-01-13 18:17:39 +08:00
# undef ARRAY_ALL
# undef ARRAY_IS_EMPTY
# undef ARRAY_JOIN
# undef ARRAY_STRINGIFY
2022-02-28 04:57:50 +08:00
p_output . append ( CLOSE_BLOCK ) ; // End of GD class.
2021-01-13 18:17:39 +08:00
}
2019-04-19 07:10:08 +08:00
void BindingsGenerator : : _generate_global_constants ( StringBuilder & p_output ) {
2017-12-24 10:17:48 +08:00
// Constants (in partial GD class)
2022-02-28 04:57:50 +08:00
p_output . append ( " namespace " BINDINGS_NAMESPACE " ; \n \n " ) ;
2019-04-19 07:10:08 +08:00
p_output . append ( " \n #pragma warning disable CS1591 // Disable warning: "
" 'Missing XML comment for publicly visible type or member' \n " ) ;
2019-03-19 04:59:35 +08:00
2022-02-28 04:57:50 +08:00
p_output . append ( " public static partial class " BINDINGS_GLOBAL_SCOPE_CLASS " \n { " ) ;
2017-12-24 10:17:48 +08:00
2021-07-16 11:45:57 +08:00
for ( const ConstantInterface & iconstant : global_constants ) {
2017-12-24 10:17:48 +08:00
if ( iconstant . const_doc & & iconstant . const_doc - > description . size ( ) ) {
2020-04-02 07:20:12 +08:00
String xml_summary = bbcode_to_xml ( fix_doc_description ( iconstant . const_doc - > description ) , nullptr ) ;
2019-04-19 07:10:08 +08:00
Vector < String > summary_lines = xml_summary . length ( ) ? xml_summary . split ( " \n " ) : Vector < String > ( ) ;
2017-12-24 10:17:48 +08:00
2019-03-19 04:59:35 +08:00
if ( summary_lines . size ( ) ) {
2019-04-19 07:10:08 +08:00
p_output . append ( MEMBER_BEGIN " /// <summary> \n " ) ;
2017-12-24 10:17:48 +08:00
2019-03-19 04:59:35 +08:00
for ( int i = 0 ; i < summary_lines . size ( ) ; i + + ) {
2022-02-28 04:57:50 +08:00
p_output . append ( INDENT1 " /// " ) ;
2019-04-19 07:10:08 +08:00
p_output . append ( summary_lines [ i ] ) ;
p_output . append ( " \n " ) ;
2017-12-24 10:17:48 +08:00
}
2022-02-28 04:57:50 +08:00
p_output . append ( INDENT1 " /// </summary> " ) ;
2019-03-19 04:59:35 +08:00
}
2017-12-24 10:17:48 +08:00
}
2022-05-09 17:47:10 +08:00
p_output . append ( MEMBER_BEGIN " public const long " ) ;
2019-04-19 07:10:08 +08:00
p_output . append ( iconstant . proxy_name ) ;
p_output . append ( " = " ) ;
p_output . append ( itos ( iconstant . value ) ) ;
p_output . append ( " ; " ) ;
2017-12-24 10:17:48 +08:00
}
2020-12-15 20:04:21 +08:00
if ( ! global_constants . is_empty ( ) ) {
2019-04-19 07:10:08 +08:00
p_output . append ( " \n " ) ;
2020-05-16 10:03:05 +08:00
}
2017-12-24 10:17:48 +08:00
2022-02-28 04:57:50 +08:00
p_output . append ( CLOSE_BLOCK ) ; // end of GD class
2017-12-24 10:17:48 +08:00
// Enums
2021-07-24 21:46:25 +08:00
for ( const EnumInterface & ienum : global_enums ) {
2020-12-15 20:04:21 +08:00
CRASH_COND ( ienum . constants . is_empty ( ) ) ;
2017-12-24 10:17:48 +08:00
String enum_proxy_name = ienum . cname . operator String ( ) ;
bool enum_in_static_class = false ;
if ( enum_proxy_name . find ( " . " ) > 0 ) {
enum_in_static_class = true ;
String enum_class_name = enum_proxy_name . get_slicec ( ' . ' , 0 ) ;
enum_proxy_name = enum_proxy_name . get_slicec ( ' . ' , 1 ) ;
CRASH_COND ( enum_class_name ! = " Variant " ) ; // Hard-coded...
2022-07-28 23:41:49 +08:00
_log ( " Declaring global enum '%s' inside struct '%s' \n " , enum_proxy_name . utf8 ( ) . get_data ( ) , enum_class_name . utf8 ( ) . get_data ( ) ) ;
2017-12-24 10:17:48 +08:00
2022-07-28 23:41:49 +08:00
p_output . append ( " \n public partial struct " ) ;
2022-12-07 23:11:39 +08:00
p_output . append ( pascal_to_pascal_case ( enum_class_name ) ) ;
2022-02-28 04:57:50 +08:00
p_output . append ( " \n " OPEN_BLOCK ) ;
2017-12-24 10:17:48 +08:00
}
2022-06-25 02:25:30 +08:00
if ( ienum . is_flags ) {
2022-02-28 04:57:50 +08:00
p_output . append ( " \n [System.Flags] " ) ;
2022-06-25 02:25:30 +08:00
}
2022-02-28 04:57:50 +08:00
p_output . append ( " \n public enum " ) ;
2022-12-07 23:11:39 +08:00
p_output . append ( pascal_to_pascal_case ( enum_proxy_name ) ) ;
2022-05-09 17:47:10 +08:00
p_output . append ( " : long " ) ;
2022-02-28 04:57:50 +08:00
p_output . append ( " \n " OPEN_BLOCK ) ;
2017-12-24 10:17:48 +08:00
2021-07-26 00:45:00 +08:00
const ConstantInterface & last = ienum . constants . back ( ) - > get ( ) ;
for ( const ConstantInterface & iconstant : ienum . constants ) {
2017-12-24 10:17:48 +08:00
if ( iconstant . const_doc & & iconstant . const_doc - > description . size ( ) ) {
2020-04-02 07:20:12 +08:00
String xml_summary = bbcode_to_xml ( fix_doc_description ( iconstant . const_doc - > description ) , nullptr ) ;
2019-04-19 07:10:08 +08:00
Vector < String > summary_lines = xml_summary . length ( ) ? xml_summary . split ( " \n " ) : Vector < String > ( ) ;
2017-12-24 10:17:48 +08:00
2019-03-19 04:59:35 +08:00
if ( summary_lines . size ( ) ) {
2022-02-28 04:57:50 +08:00
p_output . append ( INDENT1 " /// <summary> \n " ) ;
2017-12-24 10:17:48 +08:00
2019-03-19 04:59:35 +08:00
for ( int i = 0 ; i < summary_lines . size ( ) ; i + + ) {
2022-02-28 04:57:50 +08:00
p_output . append ( INDENT1 " /// " ) ;
2019-04-19 07:10:08 +08:00
p_output . append ( summary_lines [ i ] ) ;
p_output . append ( " \n " ) ;
2017-12-24 10:17:48 +08:00
}
2022-02-28 04:57:50 +08:00
p_output . append ( INDENT1 " /// </summary> \n " ) ;
2019-03-19 04:59:35 +08:00
}
2017-12-24 10:17:48 +08:00
}
2022-02-28 04:57:50 +08:00
p_output . append ( INDENT1 ) ;
2019-04-19 07:10:08 +08:00
p_output . append ( iconstant . proxy_name ) ;
p_output . append ( " = " ) ;
p_output . append ( itos ( iconstant . value ) ) ;
2021-07-26 00:45:00 +08:00
p_output . append ( & iconstant ! = & last ? " , \n " : " \n " ) ;
2017-12-24 10:17:48 +08:00
}
2022-02-28 04:57:50 +08:00
p_output . append ( CLOSE_BLOCK ) ;
2017-12-24 10:17:48 +08:00
2020-05-16 10:03:05 +08:00
if ( enum_in_static_class ) {
2022-02-28 04:57:50 +08:00
p_output . append ( CLOSE_BLOCK ) ;
2020-05-16 10:03:05 +08:00
}
2017-12-24 10:17:48 +08:00
}
2019-04-19 07:10:08 +08:00
p_output . append ( " \n #pragma warning restore CS1591 \n " ) ;
2017-12-24 10:17:48 +08:00
}
2019-12-29 02:12:32 +08:00
Error BindingsGenerator : : generate_cs_core_project ( const String & p_proj_dir ) {
2019-09-24 21:09:54 +08:00
ERR_FAIL_COND_V ( ! initialized , ERR_UNCONFIGURED ) ;
2022-03-23 17:08:58 +08:00
Ref < DirAccess > da = DirAccess : : create ( DirAccess : : ACCESS_FILESYSTEM ) ;
ERR_FAIL_COND_V ( da . is_null ( ) , ERR_CANT_CREATE ) ;
2017-10-03 05:24:00 +08:00
2019-07-03 15:44:53 +08:00
if ( ! DirAccess : : exists ( p_proj_dir ) ) {
Error err = da - > make_dir_recursive ( p_proj_dir ) ;
2019-09-25 16:28:50 +08:00
ERR_FAIL_COND_V_MSG ( err ! = OK , ERR_CANT_CREATE , " Cannot create directory ' " + p_proj_dir + " '. " ) ;
2017-10-03 05:24:00 +08:00
}
2019-07-03 15:44:53 +08:00
da - > change_dir ( p_proj_dir ) ;
2019-12-29 02:12:32 +08:00
da - > make_dir ( " Generated " ) ;
da - > make_dir ( " Generated/GodotObjects " ) ;
2017-10-03 05:24:00 +08:00
2019-12-29 02:12:32 +08:00
String base_gen_dir = path : : join ( p_proj_dir , " Generated " ) ;
String godot_objects_gen_dir = path : : join ( base_gen_dir , " GodotObjects " ) ;
Vector < String > compile_items ;
2017-10-03 05:24:00 +08:00
2017-12-24 10:17:48 +08:00
// Generate source file for global scope constants and enums
{
2019-04-19 07:10:08 +08:00
StringBuilder constants_source ;
2017-12-24 10:17:48 +08:00
_generate_global_constants ( constants_source ) ;
2019-12-29 02:12:32 +08:00
String output_file = path : : join ( base_gen_dir , BINDINGS_GLOBAL_SCOPE_CLASS " _constants.cs " ) ;
2017-12-24 10:17:48 +08:00
Error save_err = _save_file ( output_file , constants_source ) ;
2020-05-16 10:03:05 +08:00
if ( save_err ! = OK ) {
2017-12-24 10:17:48 +08:00
return save_err ;
2020-05-16 10:03:05 +08:00
}
2017-12-24 10:17:48 +08:00
2019-12-29 02:12:32 +08:00
compile_items . push_back ( output_file ) ;
2017-12-24 10:17:48 +08:00
}
2021-01-13 18:17:39 +08:00
// Generate source file for array extensions
{
StringBuilder extensions_source ;
_generate_array_extensions ( extensions_source ) ;
String output_file = path : : join ( base_gen_dir , BINDINGS_GLOBAL_SCOPE_CLASS " _extensions.cs " ) ;
Error save_err = _save_file ( output_file , extensions_source ) ;
if ( save_err ! = OK ) {
return save_err ;
}
2019-12-29 02:12:32 +08:00
compile_items . push_back ( output_file ) ;
2017-12-24 10:17:48 +08:00
}
2022-05-08 16:09:19 +08:00
for ( const KeyValue < StringName , TypeInterface > & E : obj_types ) {
const TypeInterface & itype = E . value ;
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
if ( itype . api_type = = ClassDB : : API_EDITOR ) {
2017-10-03 05:24:00 +08:00
continue ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
2019-12-29 02:12:32 +08:00
String output_file = path : : join ( godot_objects_gen_dir , itype . proxy_name + " .cs " ) ;
2018-01-10 02:06:59 +08:00
Error err = _generate_cs_type ( itype , output_file ) ;
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
if ( err = = ERR_SKIP ) {
2017-10-03 05:24:00 +08:00
continue ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
if ( err ! = OK ) {
2017-10-03 05:24:00 +08:00
return err ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
2019-12-29 02:12:32 +08:00
compile_items . push_back ( output_file ) ;
2017-10-03 05:24:00 +08:00
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
// Generate native calls
2017-10-03 05:24:00 +08:00
2019-04-19 07:10:08 +08:00
StringBuilder cs_icalls_content ;
2022-02-28 04:57:50 +08:00
cs_icalls_content . append ( " namespace " BINDINGS_NAMESPACE " ; \n \n " ) ;
2019-04-19 07:10:08 +08:00
cs_icalls_content . append ( " using System; \n "
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
" using System.Diagnostics.CodeAnalysis; \n "
" using System.Runtime.InteropServices; \n "
" using Godot.NativeInterop; \n "
2019-04-19 07:10:08 +08:00
" \n " ) ;
2022-02-28 04:57:50 +08:00
cs_icalls_content . append ( " [SuppressMessage( \" ReSharper \" , \" InconsistentNaming \" )] \n " ) ;
cs_icalls_content . append ( " [SuppressMessage( \" ReSharper \" , \" RedundantUnsafeContext \" )] \n " ) ;
cs_icalls_content . append ( " [SuppressMessage( \" ReSharper \" , \" RedundantNameQualifier \" )] \n " ) ;
cs_icalls_content . append ( " [System.Runtime.CompilerServices.SkipLocalsInit] \n " ) ;
cs_icalls_content . append ( " internal static class " BINDINGS_CLASS_NATIVECALLS " \n { " ) ;
2019-04-19 07:10:08 +08:00
cs_icalls_content . append ( MEMBER_BEGIN " internal static ulong godot_api_hash = " ) ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
cs_icalls_content . append ( String : : num_uint64 ( ClassDB : : get_api_hash ( ClassDB : : API_CORE ) ) + " ; \n " ) ;
2019-04-19 07:10:08 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
cs_icalls_content . append ( MEMBER_BEGIN " private const int VarArgsSpanThreshold = 10; \n " ) ;
2017-10-03 05:24:00 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
for ( const InternalCall & icall : method_icalls ) {
if ( icall . editor_only ) {
continue ;
}
Error err = _generate_cs_native_calls ( icall , cs_icalls_content ) ;
if ( err ! = OK ) {
return err ;
}
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
2022-02-28 04:57:50 +08:00
cs_icalls_content . append ( CLOSE_BLOCK ) ;
2017-10-03 05:24:00 +08:00
2019-12-29 02:12:32 +08:00
String internal_methods_file = path : : join ( base_gen_dir , BINDINGS_CLASS_NATIVECALLS " .cs " ) ;
2017-10-03 05:24:00 +08:00
Error err = _save_file ( internal_methods_file , cs_icalls_content ) ;
2020-05-16 10:03:05 +08:00
if ( err ! = OK ) {
2017-10-03 05:24:00 +08:00
return err ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
2019-12-29 02:12:32 +08:00
compile_items . push_back ( internal_methods_file ) ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
// Generate GeneratedIncludes.props
2019-12-29 02:12:32 +08:00
StringBuilder includes_props_content ;
includes_props_content . append ( " <Project> \n "
" <ItemGroup> \n " ) ;
for ( int i = 0 ; i < compile_items . size ( ) ; i + + ) {
String include = path : : relative_to ( compile_items [ i ] , p_proj_dir ) . replace ( " / " , " \\ " ) ;
includes_props_content . append ( " <Compile Include= \" " + include + " \" /> \n " ) ;
}
includes_props_content . append ( " </ItemGroup> \n "
" </Project> \n " ) ;
String includes_props_file = path : : join ( base_gen_dir , " GeneratedIncludes.props " ) ;
err = _save_file ( includes_props_file , includes_props_content ) ;
2020-05-16 10:03:05 +08:00
if ( err ! = OK ) {
2019-12-29 02:12:32 +08:00
return err ;
2020-05-16 10:03:05 +08:00
}
2017-10-24 22:18:47 +08:00
2017-10-03 05:24:00 +08:00
return OK ;
}
2019-12-29 02:12:32 +08:00
Error BindingsGenerator : : generate_cs_editor_project ( const String & p_proj_dir ) {
2019-09-24 21:09:54 +08:00
ERR_FAIL_COND_V ( ! initialized , ERR_UNCONFIGURED ) ;
2022-03-23 17:08:58 +08:00
Ref < DirAccess > da = DirAccess : : create ( DirAccess : : ACCESS_FILESYSTEM ) ;
ERR_FAIL_COND_V ( da . is_null ( ) , ERR_CANT_CREATE ) ;
2017-10-03 05:24:00 +08:00
2019-07-03 15:44:53 +08:00
if ( ! DirAccess : : exists ( p_proj_dir ) ) {
Error err = da - > make_dir_recursive ( p_proj_dir ) ;
2017-10-03 05:24:00 +08:00
ERR_FAIL_COND_V ( err ! = OK , ERR_CANT_CREATE ) ;
}
2019-07-03 15:44:53 +08:00
da - > change_dir ( p_proj_dir ) ;
2019-12-29 02:12:32 +08:00
da - > make_dir ( " Generated " ) ;
da - > make_dir ( " Generated/GodotObjects " ) ;
String base_gen_dir = path : : join ( p_proj_dir , " Generated " ) ;
String godot_objects_gen_dir = path : : join ( base_gen_dir , " GodotObjects " ) ;
2017-10-03 05:24:00 +08:00
2019-12-29 02:12:32 +08:00
Vector < String > compile_items ;
2017-10-03 05:24:00 +08:00
2022-05-08 16:09:19 +08:00
for ( const KeyValue < StringName , TypeInterface > & E : obj_types ) {
const TypeInterface & itype = E . value ;
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
if ( itype . api_type ! = ClassDB : : API_EDITOR ) {
2017-10-03 05:24:00 +08:00
continue ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
2019-12-29 02:12:32 +08:00
String output_file = path : : join ( godot_objects_gen_dir , itype . proxy_name + " .cs " ) ;
2018-01-10 02:06:59 +08:00
Error err = _generate_cs_type ( itype , output_file ) ;
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
if ( err = = ERR_SKIP ) {
2017-10-03 05:24:00 +08:00
continue ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
if ( err ! = OK ) {
2017-10-03 05:24:00 +08:00
return err ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
2019-12-29 02:12:32 +08:00
compile_items . push_back ( output_file ) ;
2017-10-03 05:24:00 +08:00
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
// Generate native calls
2019-04-19 07:10:08 +08:00
StringBuilder cs_icalls_content ;
2022-02-28 04:57:50 +08:00
cs_icalls_content . append ( " namespace " BINDINGS_NAMESPACE " ; \n \n " ) ;
2019-04-19 07:10:08 +08:00
cs_icalls_content . append ( " using System; \n "
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
" using System.Diagnostics.CodeAnalysis; \n "
" using System.Runtime.InteropServices; \n "
" using Godot.NativeInterop; \n "
2019-04-19 07:10:08 +08:00
" \n " ) ;
2022-02-28 04:57:50 +08:00
cs_icalls_content . append ( " [SuppressMessage( \" ReSharper \" , \" InconsistentNaming \" )] \n " ) ;
cs_icalls_content . append ( " [SuppressMessage( \" ReSharper \" , \" RedundantUnsafeContext \" )] \n " ) ;
cs_icalls_content . append ( " [SuppressMessage( \" ReSharper \" , \" RedundantNameQualifier \" )] \n " ) ;
cs_icalls_content . append ( " [System.Runtime.CompilerServices.SkipLocalsInit] \n " ) ;
cs_icalls_content . append ( " internal static class " BINDINGS_CLASS_NATIVECALLS_EDITOR " \n " OPEN_BLOCK ) ;
cs_icalls_content . append ( INDENT1 " internal static ulong godot_api_hash = " ) ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
cs_icalls_content . append ( String : : num_uint64 ( ClassDB : : get_api_hash ( ClassDB : : API_EDITOR ) ) + " ; \n " ) ;
2019-04-19 07:10:08 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
cs_icalls_content . append ( MEMBER_BEGIN " private const int VarArgsSpanThreshold = 10; \n " ) ;
2017-10-03 05:24:00 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
cs_icalls_content . append ( " \n " ) ;
2017-10-03 05:24:00 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
for ( const InternalCall & icall : method_icalls ) {
if ( ! icall . editor_only ) {
continue ;
}
Error err = _generate_cs_native_calls ( icall , cs_icalls_content ) ;
if ( err ! = OK ) {
return err ;
}
}
2017-10-03 05:24:00 +08:00
2022-02-28 04:57:50 +08:00
cs_icalls_content . append ( CLOSE_BLOCK ) ;
2017-10-03 05:24:00 +08:00
2019-12-29 02:12:32 +08:00
String internal_methods_file = path : : join ( base_gen_dir , BINDINGS_CLASS_NATIVECALLS_EDITOR " .cs " ) ;
2017-10-03 05:24:00 +08:00
Error err = _save_file ( internal_methods_file , cs_icalls_content ) ;
2020-05-16 10:03:05 +08:00
if ( err ! = OK ) {
2017-10-03 05:24:00 +08:00
return err ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
2019-12-29 02:12:32 +08:00
compile_items . push_back ( internal_methods_file ) ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
// Generate GeneratedIncludes.props
2019-12-29 02:12:32 +08:00
StringBuilder includes_props_content ;
includes_props_content . append ( " <Project> \n "
" <ItemGroup> \n " ) ;
for ( int i = 0 ; i < compile_items . size ( ) ; i + + ) {
String include = path : : relative_to ( compile_items [ i ] , p_proj_dir ) . replace ( " / " , " \\ " ) ;
includes_props_content . append ( " <Compile Include= \" " + include + " \" /> \n " ) ;
}
includes_props_content . append ( " </ItemGroup> \n "
" </Project> \n " ) ;
String includes_props_file = path : : join ( base_gen_dir , " GeneratedIncludes.props " ) ;
err = _save_file ( includes_props_file , includes_props_content ) ;
2020-05-16 10:03:05 +08:00
if ( err ! = OK ) {
2019-12-29 02:12:32 +08:00
return err ;
2020-05-16 10:03:05 +08:00
}
2018-11-08 08:05:19 +08:00
return OK ;
}
2017-10-03 05:24:00 +08:00
2019-04-25 21:45:12 +08:00
Error BindingsGenerator : : generate_cs_api ( const String & p_output_dir ) {
2019-09-24 21:09:54 +08:00
ERR_FAIL_COND_V ( ! initialized , ERR_UNCONFIGURED ) ;
2019-07-08 21:16:35 +08:00
String output_dir = path : : abspath ( path : : realpath ( p_output_dir ) ) ;
2019-07-03 15:44:53 +08:00
2022-03-23 17:08:58 +08:00
Ref < DirAccess > da = DirAccess : : create ( DirAccess : : ACCESS_FILESYSTEM ) ;
ERR_FAIL_COND_V ( da . is_null ( ) , ERR_CANT_CREATE ) ;
2018-11-08 08:05:19 +08:00
2019-07-03 15:44:53 +08:00
if ( ! DirAccess : : exists ( output_dir ) ) {
Error err = da - > make_dir_recursive ( output_dir ) ;
2018-11-08 08:05:19 +08:00
ERR_FAIL_COND_V ( err ! = OK , ERR_CANT_CREATE ) ;
}
2019-07-03 15:44:53 +08:00
Error proj_err ;
2018-11-08 08:05:19 +08:00
2019-07-03 15:44:53 +08:00
// Generate GodotSharp source files
2018-11-08 08:05:19 +08:00
2022-08-30 08:34:01 +08:00
String core_proj_dir = output_dir . path_join ( CORE_API_ASSEMBLY_NAME ) ;
2018-11-08 08:05:19 +08:00
2019-12-29 02:12:32 +08:00
proj_err = generate_cs_core_project ( core_proj_dir ) ;
2018-11-08 08:05:19 +08:00
if ( proj_err ! = OK ) {
2019-08-09 09:39:45 +08:00
ERR_PRINT ( " Generation of the Core API C# project failed. " ) ;
2018-11-08 08:05:19 +08:00
return proj_err ;
}
2019-07-03 15:44:53 +08:00
// Generate GodotSharpEditor source files
2022-08-30 08:34:01 +08:00
String editor_proj_dir = output_dir . path_join ( EDITOR_API_ASSEMBLY_NAME ) ;
2019-07-03 15:44:53 +08:00
2019-12-29 02:12:32 +08:00
proj_err = generate_cs_editor_project ( editor_proj_dir ) ;
2018-11-08 08:05:19 +08:00
if ( proj_err ! = OK ) {
2019-08-09 09:39:45 +08:00
ERR_PRINT ( " Generation of the Editor API C# project failed. " ) ;
2018-11-08 08:05:19 +08:00
return proj_err ;
}
2017-10-03 05:24:00 +08:00
2019-12-29 02:12:32 +08:00
_log ( " The Godot API sources were successfully generated \n " ) ;
2019-07-03 15:44:53 +08:00
2017-10-03 05:24:00 +08:00
return OK ;
}
2018-09-04 11:40:41 +08:00
// FIXME: There are some members that hide other inherited members.
// - In the case of both members being the same kind, the new one must be declared
2019-08-09 09:39:45 +08:00
// explicitly as 'new' to avoid the warning (and we must print a message about it).
2018-09-04 11:40:41 +08:00
// - In the case of both members being of a different kind, then the new one must
// be renamed to avoid the name collision (and we must print a warning about it).
// - Csc warning e.g.:
// ObjectType/LineEdit.cs(140,38): warning CS0108: 'LineEdit.FocusMode' hides inherited member 'Control.FocusMode'. Use the new keyword if hiding was intended.
2017-10-03 05:24:00 +08:00
Error BindingsGenerator : : _generate_cs_type ( const TypeInterface & itype , const String & p_output_file ) {
2018-09-04 11:40:41 +08:00
CRASH_COND ( ! itype . is_object_type ) ;
2017-12-24 10:17:48 +08:00
bool is_derived_type = itype . base_name ! = StringName ( ) ;
2017-10-03 05:24:00 +08:00
2018-09-04 11:40:41 +08:00
if ( ! is_derived_type ) {
2022-12-07 23:16:51 +08:00
// Some GodotObject assertions
2018-09-04 11:40:41 +08:00
CRASH_COND ( itype . cname ! = name_cache . type_Object ) ;
CRASH_COND ( ! itype . is_instantiable ) ;
CRASH_COND ( itype . api_type ! = ClassDB : : API_CORE ) ;
2021-06-05 00:03:15 +08:00
CRASH_COND ( itype . is_ref_counted ) ;
2018-09-04 11:40:41 +08:00
CRASH_COND ( itype . is_singleton ) ;
}
2019-04-25 21:45:12 +08:00
_log ( " Generating %s.cs... \n " , itype . proxy_name . utf8 ( ) . get_data ( ) ) ;
2017-10-03 05:24:00 +08:00
2019-04-19 07:10:08 +08:00
StringBuilder output ;
2017-10-03 05:24:00 +08:00
2022-02-28 04:57:50 +08:00
output . append ( " namespace " BINDINGS_NAMESPACE " ; \n \n " ) ;
2019-04-19 07:10:08 +08:00
output . append ( " using System; \n " ) ; // IntPtr
output . append ( " using System.Diagnostics; \n " ) ; // DebuggerBrowsable
2021-09-13 02:21:15 +08:00
output . append ( " using Godot.NativeInterop; \n " ) ;
2018-10-16 23:22:27 +08:00
2019-04-19 07:10:08 +08:00
output . append ( " \n "
" #pragma warning disable CS1591 // Disable warning: "
" 'Missing XML comment for publicly visible type or member' \n "
" #pragma warning disable CS1573 // Disable warning: "
" 'Parameter has no matching param tag in the XML comment' \n " ) ;
2018-10-16 23:22:27 +08:00
2021-12-29 06:25:16 +08:00
output . append ( " \n #nullable disable \n " ) ;
2017-10-03 05:24:00 +08:00
const DocData : : ClassDoc * class_doc = itype . class_doc ;
if ( class_doc & & class_doc - > description . size ( ) ) {
2019-03-19 04:59:35 +08:00
String xml_summary = bbcode_to_xml ( fix_doc_description ( class_doc - > description ) , & itype ) ;
2019-04-19 07:10:08 +08:00
Vector < String > summary_lines = xml_summary . length ( ) ? xml_summary . split ( " \n " ) : Vector < String > ( ) ;
2017-10-03 05:24:00 +08:00
2019-03-19 04:59:35 +08:00
if ( summary_lines . size ( ) ) {
2022-02-28 04:57:50 +08:00
output . append ( " /// <summary> \n " ) ;
2017-10-03 05:24:00 +08:00
2019-03-19 04:59:35 +08:00
for ( int i = 0 ; i < summary_lines . size ( ) ; i + + ) {
2022-02-28 04:57:50 +08:00
output . append ( " /// " ) ;
2019-04-19 07:10:08 +08:00
output . append ( summary_lines [ i ] ) ;
output . append ( " \n " ) ;
2017-10-03 05:24:00 +08:00
}
2022-02-28 04:57:50 +08:00
output . append ( " /// </summary> \n " ) ;
2019-03-19 04:59:35 +08:00
}
2017-10-03 05:24:00 +08:00
}
2022-02-28 04:57:30 +08:00
// We generate a `GodotClassName` attribute if the engine class name is not the same as the
// generated C# class name. This allows introspection code to find the name associated with
// the class. If the attribute is not present, the C# class name can be used instead.
if ( itype . name ! = itype . proxy_name ) {
2022-02-28 04:57:50 +08:00
output < < " [GodotClassName( \" " < < itype . name < < " \" )] \n " ;
2022-02-28 04:57:30 +08:00
}
2022-02-28 04:57:50 +08:00
output . append ( " public " ) ;
2018-09-04 11:40:41 +08:00
if ( itype . is_singleton ) {
2019-04-19 07:10:08 +08:00
output . append ( " static partial class " ) ;
2018-08-28 02:39:51 +08:00
} else {
2021-09-13 02:21:15 +08:00
// Even if the class is not instantiable, we can't declare it abstract because
// the engine can still instantiate them and return them via the scripting API.
// Example: `SceneTreeTimer` returned from `SceneTree.create_timer`.
// See the reverted commit: ef5672d3f94a7321ed779c922088bb72adbb1521
output . append ( " partial class " ) ;
2018-08-28 02:39:51 +08:00
}
2019-04-19 07:10:08 +08:00
output . append ( itype . proxy_name ) ;
2017-10-03 05:24:00 +08:00
2021-12-29 06:25:16 +08:00
if ( is_derived_type & & ! itype . is_singleton ) {
2018-09-04 11:40:41 +08:00
if ( obj_types . has ( itype . base_name ) ) {
2019-04-19 07:10:08 +08:00
output . append ( " : " ) ;
output . append ( obj_types [ itype . base_name ] . proxy_name ) ;
2018-09-04 11:40:41 +08:00
} else {
2019-11-07 00:03:04 +08:00
ERR_PRINT ( " Base type ' " + itype . base_name . operator String ( ) + " ' does not exist, for class ' " + itype . name + " '. " ) ;
2018-09-04 11:40:41 +08:00
return ERR_INVALID_DATA ;
}
2017-10-03 05:24:00 +08:00
}
2022-02-28 04:57:50 +08:00
output . append ( " \n { " ) ;
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
// Add constants
2017-10-03 05:24:00 +08:00
2021-07-16 11:45:57 +08:00
for ( const ConstantInterface & iconstant : itype . constants ) {
2020-05-16 10:03:05 +08:00
if ( iconstant . const_doc & & iconstant . const_doc - > description . size ( ) ) {
String xml_summary = bbcode_to_xml ( fix_doc_description ( iconstant . const_doc - > description ) , & itype ) ;
Vector < String > summary_lines = xml_summary . length ( ) ? xml_summary . split ( " \n " ) : Vector < String > ( ) ;
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
if ( summary_lines . size ( ) ) {
output . append ( MEMBER_BEGIN " /// <summary> \n " ) ;
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
for ( int i = 0 ; i < summary_lines . size ( ) ; i + + ) {
2022-02-28 04:57:50 +08:00
output . append ( INDENT1 " /// " ) ;
2020-05-16 10:03:05 +08:00
output . append ( summary_lines [ i ] ) ;
output . append ( " \n " ) ;
2019-03-19 04:59:35 +08:00
}
2017-10-03 05:24:00 +08:00
2022-02-28 04:57:50 +08:00
output . append ( INDENT1 " /// </summary> " ) ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
}
2022-05-09 17:47:10 +08:00
output . append ( MEMBER_BEGIN " public const long " ) ;
2020-05-16 10:03:05 +08:00
output . append ( iconstant . proxy_name ) ;
output . append ( " = " ) ;
output . append ( itos ( iconstant . value ) ) ;
output . append ( " ; " ) ;
}
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
if ( itype . constants . size ( ) ) {
output . append ( " \n " ) ;
}
2017-12-24 10:17:48 +08:00
2020-05-16 10:03:05 +08:00
// Add enums
2017-12-24 10:17:48 +08:00
2021-07-16 11:45:57 +08:00
for ( const EnumInterface & ienum : itype . enums ) {
2020-12-15 20:04:21 +08:00
ERR_FAIL_COND_V ( ienum . constants . is_empty ( ) , ERR_BUG ) ;
2017-12-24 10:17:48 +08:00
2022-06-25 02:25:30 +08:00
if ( ienum . is_flags ) {
output . append ( MEMBER_BEGIN " [System.Flags] " ) ;
}
2020-05-16 10:03:05 +08:00
output . append ( MEMBER_BEGIN " public enum " ) ;
2022-12-07 23:11:39 +08:00
output . append ( pascal_to_pascal_case ( ienum . cname . operator String ( ) ) ) ;
2022-05-09 17:47:10 +08:00
output . append ( " : long " ) ;
2020-05-16 10:03:05 +08:00
output . append ( MEMBER_BEGIN OPEN_BLOCK ) ;
2017-12-24 10:17:48 +08:00
2021-07-26 00:45:00 +08:00
const ConstantInterface & last = ienum . constants . back ( ) - > get ( ) ;
for ( const ConstantInterface & iconstant : ienum . constants ) {
2020-05-16 10:03:05 +08:00
if ( iconstant . const_doc & & iconstant . const_doc - > description . size ( ) ) {
String xml_summary = bbcode_to_xml ( fix_doc_description ( iconstant . const_doc - > description ) , & itype ) ;
Vector < String > summary_lines = xml_summary . length ( ) ? xml_summary . split ( " \n " ) : Vector < String > ( ) ;
2017-12-24 10:17:48 +08:00
2020-05-16 10:03:05 +08:00
if ( summary_lines . size ( ) ) {
2022-02-28 04:57:50 +08:00
output . append ( INDENT2 " /// <summary> \n " ) ;
2017-12-24 10:17:48 +08:00
2020-05-16 10:03:05 +08:00
for ( int i = 0 ; i < summary_lines . size ( ) ; i + + ) {
2022-02-28 04:57:50 +08:00
output . append ( INDENT2 " /// " ) ;
2020-05-16 10:03:05 +08:00
output . append ( summary_lines [ i ] ) ;
output . append ( " \n " ) ;
2019-03-19 04:59:35 +08:00
}
2017-12-24 10:17:48 +08:00
2022-02-28 04:57:50 +08:00
output . append ( INDENT2 " /// </summary> \n " ) ;
2020-05-16 10:03:05 +08:00
}
2017-12-24 10:17:48 +08:00
}
2022-02-28 04:57:50 +08:00
output . append ( INDENT2 ) ;
2020-05-16 10:03:05 +08:00
output . append ( iconstant . proxy_name ) ;
output . append ( " = " ) ;
output . append ( itos ( iconstant . value ) ) ;
2021-07-26 00:45:00 +08:00
output . append ( & iconstant ! = & last ? " , \n " : " \n " ) ;
2017-12-24 10:17:48 +08:00
}
2022-02-28 04:57:50 +08:00
output . append ( INDENT1 CLOSE_BLOCK ) ;
2017-10-03 05:24:00 +08:00
}
2020-05-16 10:03:05 +08:00
// Add properties
2021-07-16 11:45:57 +08:00
for ( const PropertyInterface & iprop : itype . properties ) {
2020-05-16 10:03:05 +08:00
Error prop_err = _generate_cs_property ( itype , iprop , output ) ;
ERR_FAIL_COND_V_MSG ( prop_err ! = OK , prop_err ,
" Failed to generate property ' " + iprop . cname . operator String ( ) +
" ' for class ' " + itype . name + " '. " ) ;
}
2019-02-05 03:39:02 +08:00
2018-09-04 11:40:41 +08:00
if ( itype . is_singleton ) {
2017-10-03 05:24:00 +08:00
// Add the type name and the singleton pointer as static fields
2022-12-07 23:16:51 +08:00
output . append ( MEMBER_BEGIN " private static GodotObject singleton; \n " ) ;
2019-04-19 07:10:08 +08:00
2022-12-07 23:16:51 +08:00
output < < MEMBER_BEGIN " public static GodotObject " CS_PROPERTY_SINGLETON " \n " INDENT1 " { \n "
2022-02-28 04:57:50 +08:00
< < INDENT2 " get \n " INDENT2 " { \n " INDENT3 " if (singleton == null) \n "
2023-03-03 17:05:05 +08:00
< < INDENT4 " singleton = " C_METHOD_ENGINE_GET_SINGLETON " ( \" "
< < itype . name
< < " \" ); \n " INDENT3 " return singleton; \n " INDENT2 " } \n " INDENT1 " } \n " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
output . append ( MEMBER_BEGIN " private static readonly StringName " BINDINGS_NATIVE_NAME_FIELD " = \" " ) ;
2019-04-19 07:10:08 +08:00
output . append ( itype . name ) ;
output . append ( " \" ; \n " ) ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
} else {
2022-12-07 23:16:51 +08:00
// IMPORTANT: We also generate the static fields for GodotObject instead of declaring
// them manually in the `GodotObject.base.cs` partial class declaration, because they're
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
// required by other static fields in this generated partial class declaration.
// Static fields are initialized in order of declaration, but when they're in different
// partial class declarations then it becomes harder to tell (Rider warns about this).
2019-04-19 07:10:08 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
// Add native name static field
2017-10-03 05:24:00 +08:00
2021-09-13 02:21:15 +08:00
if ( is_derived_type ) {
2021-12-29 06:25:16 +08:00
output < < MEMBER_BEGIN " private static readonly System.Type CachedType = typeof( " < < itype . proxy_name < < " ); \n " ;
2021-09-13 02:21:15 +08:00
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
output . append ( MEMBER_BEGIN " private static readonly StringName " BINDINGS_NATIVE_NAME_FIELD " = \" " ) ;
2019-04-19 07:10:08 +08:00
output . append ( itype . name ) ;
output . append ( " \" ; \n " ) ;
2017-10-03 05:24:00 +08:00
if ( itype . is_instantiable ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
// Add native constructor static field
output < < MEMBER_BEGIN < < " [DebuggerBrowsable(DebuggerBrowsableState.Never)] \n "
2022-02-28 04:57:50 +08:00
< < INDENT1 " private static readonly unsafe delegate* unmanaged<IntPtr> "
2021-09-13 02:21:15 +08:00
< < CS_STATIC_FIELD_NATIVE_CTOR " = " ICALL_CLASSDB_GET_CONSTRUCTOR
2021-09-13 02:23:05 +08:00
< < " ( " BINDINGS_NATIVE_NAME_FIELD " ); \n " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
}
if ( is_derived_type ) {
// Add default constructor
if ( itype . is_instantiable ) {
output < < MEMBER_BEGIN " public " < < itype . proxy_name < < " () : this( "
2022-02-28 04:57:50 +08:00
< < ( itype . memory_own ? " true " : " false " ) < < " ) \n " OPEN_BLOCK_L1
< < INDENT2 " unsafe \n " INDENT2 OPEN_BLOCK
< < INDENT3 " _ConstructAndInitialize( " CS_STATIC_FIELD_NATIVE_CTOR " , "
2021-12-29 06:25:16 +08:00
< < BINDINGS_NATIVE_NAME_FIELD " , CachedType, refCounted: "
< < ( itype . is_ref_counted ? " true " : " false " ) < < " ); \n "
2022-02-28 04:57:50 +08:00
< < CLOSE_BLOCK_L2 CLOSE_BLOCK_L1 ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
} else {
// Hide the constructor
output . append ( MEMBER_BEGIN " internal " ) ;
output . append ( itype . proxy_name ) ;
output . append ( " () {} \n " ) ;
}
// Add.. em.. trick constructor. Sort of.
2019-04-19 07:10:08 +08:00
output . append ( MEMBER_BEGIN " internal " ) ;
output . append ( itype . proxy_name ) ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
output . append ( " (bool " CS_PARAM_MEMORYOWN " ) : base( " CS_PARAM_MEMORYOWN " ) {} \n " ) ;
2017-10-03 05:24:00 +08:00
}
}
2021-09-13 02:21:15 +08:00
// Methods
2017-10-29 09:37:13 +08:00
int method_bind_count = 0 ;
2021-07-16 11:45:57 +08:00
for ( const MethodInterface & imethod : itype . methods ) {
2017-10-29 09:37:13 +08:00
Error method_err = _generate_cs_method ( itype , imethod , method_bind_count , output ) ;
2019-08-09 09:39:45 +08:00
ERR_FAIL_COND_V_MSG ( method_err ! = OK , method_err ,
" Failed to generate method ' " + imethod . name + " ' for class ' " + itype . name + " '. " ) ;
2017-10-29 09:37:13 +08:00
}
2017-10-03 05:24:00 +08:00
2021-09-13 02:21:15 +08:00
// Signals
2021-07-16 11:45:57 +08:00
for ( const SignalInterface & isignal : itype . signals_ ) {
2020-03-15 02:20:17 +08:00
Error method_err = _generate_cs_signal ( itype , isignal , output ) ;
ERR_FAIL_COND_V_MSG ( method_err ! = OK , method_err ,
" Failed to generate signal ' " + isignal . name + " ' for class ' " + itype . name + " '. " ) ;
}
2021-12-29 06:25:16 +08:00
// Script members look-up
2021-09-13 02:21:15 +08:00
if ( ! itype . is_singleton & & ( is_derived_type | | itype . has_virtual_methods ) ) {
2021-12-29 06:25:16 +08:00
// Generate method names cache fields
2021-12-29 06:25:16 +08:00
for ( const MethodInterface & imethod : itype . methods ) {
if ( ! imethod . is_virtual ) {
continue ;
}
output < < MEMBER_BEGIN " // ReSharper disable once InconsistentNaming \n "
2022-02-28 04:57:50 +08:00
< < INDENT1 " [DebuggerBrowsable(DebuggerBrowsableState.Never)] \n "
< < INDENT1 " private static readonly StringName "
2021-12-29 06:25:16 +08:00
< < CS_STATIC_FIELD_METHOD_PROXY_NAME_PREFIX < < imethod . name
< < " = \" " < < imethod . proxy_name < < " \" ; \n " ;
}
2023-01-14 03:59:02 +08:00
// Generate signal names cache fields
for ( const SignalInterface & isignal : itype . signals_ ) {
output < < MEMBER_BEGIN " // ReSharper disable once InconsistentNaming \n "
< < INDENT1 " [DebuggerBrowsable(DebuggerBrowsableState.Never)] \n "
< < INDENT1 " private static readonly StringName "
< < CS_STATIC_FIELD_SIGNAL_PROXY_NAME_PREFIX < < isignal . name
< < " = \" " < < isignal . proxy_name < < " \" ; \n " ;
}
2021-12-29 06:25:16 +08:00
// TODO: Only generate HasGodotClassMethod and InvokeGodotClassMethod if there's any method
// Generate InvokeGodotClassMethod
2021-12-29 06:25:16 +08:00
output < < MEMBER_BEGIN " protected internal " < < ( is_derived_type ? " override " : " virtual " )
< < " bool " CS_METHOD_INVOKE_GODOT_CLASS_METHOD " (in godot_string_name method, "
C#: Remove need for reflection to invoking callable delegates
We aim to make the C# API reflection-free, mainly for concerns about
performance, and to be able to target NativeAOT in refletion-free mode,
which reduces the binary size.
One of the main usages of reflection still left was the dynamic
invokation of callable delegates, and for some time I wasn't sure
I would find an alternative solution that I'd be happy with.
The new solution uses trampoline functions to invoke the delegates:
```
static void Trampoline(object delegateObj, NativeVariantPtrArgs args, out godot_variant ret)
{
if (args.Count != 1)
throw new ArgumentException($"Callable expected 1 arguments but received {args.Count}.");
string res = ((Func<int, string>)delegateObj)(
VariantConversionCallbacks.GetToManagedCallback<int>()(args[0])
);
ret = VariantConversionCallbacks.GetToVariantCallback<string>()(res);
}
Callable.CreateWithUnsafeTrampoline((int num) => "Foo" + num, &Trampoline);
```
Of course, this is too much boilerplate for user code. To improve this,
the `Callable.From` methods were added. These are overloads that take
`Action` and `Func` delegates, which covers the most common use cases:
lambdas and method groups:
```
// Lambda
Callable.From((int num) => "Foo" + num);
// Method group
string AppendNum(int num) => "Foo" + num;
Callable.From(AppendNum);
```
Unfortunately, due to limitations in the C# language, implicit
conversions from delegates to `Callable` are not supported.
`Callable.From` does not support custom delegates. These should be
uncommon, but the Godot C# API actually uses them for event signals.
As such, the bindings generator was updated to generate trampoline
functions for event signals. It was also optimized to use `Action`
instead of a custom delegate for parameterless signals, which removes
the need for the trampoline functions for those signals.
The change to reflection-free invokation removes one of the last needs
for `ConvertVariantToManagedObjectOfType`. The only remaining usage is
from calling script constructors with parameters from the engine
(`CreateManagedForGodotObjectScriptInstance`). Once that one is made
reflection-free, `ConvertVariantToManagedObjectOfType` can be removed.
2022-10-29 04:59:13 +08:00
< < " NativeVariantPtrArgs args, out godot_variant ret) \n "
2022-02-28 04:57:50 +08:00
< < INDENT1 " { \n " ;
2021-09-13 02:21:15 +08:00
for ( const MethodInterface & imethod : itype . methods ) {
if ( ! imethod . is_virtual ) {
continue ;
}
2021-12-29 06:25:16 +08:00
// We also call HasGodotClassMethod to ensure the method is overridden and avoid calling
// the stub implementation. This solution adds some extra overhead to calls, but it's
// much simpler than other solutions. This won't be a problem once we move to function
// pointers of generated wrappers for each method, as lookup will only happen once.
// We check both native names (snake_case) and proxy names (PascalCase)
2022-02-28 04:57:50 +08:00
output < < INDENT2 " if ((method == " < < CS_STATIC_FIELD_METHOD_PROXY_NAME_PREFIX < < imethod . name
2022-08-25 23:07:20 +08:00
< < " || method == MethodName. " < < imethod . proxy_name
C#: Remove need for reflection to invoking callable delegates
We aim to make the C# API reflection-free, mainly for concerns about
performance, and to be able to target NativeAOT in refletion-free mode,
which reduces the binary size.
One of the main usages of reflection still left was the dynamic
invokation of callable delegates, and for some time I wasn't sure
I would find an alternative solution that I'd be happy with.
The new solution uses trampoline functions to invoke the delegates:
```
static void Trampoline(object delegateObj, NativeVariantPtrArgs args, out godot_variant ret)
{
if (args.Count != 1)
throw new ArgumentException($"Callable expected 1 arguments but received {args.Count}.");
string res = ((Func<int, string>)delegateObj)(
VariantConversionCallbacks.GetToManagedCallback<int>()(args[0])
);
ret = VariantConversionCallbacks.GetToVariantCallback<string>()(res);
}
Callable.CreateWithUnsafeTrampoline((int num) => "Foo" + num, &Trampoline);
```
Of course, this is too much boilerplate for user code. To improve this,
the `Callable.From` methods were added. These are overloads that take
`Action` and `Func` delegates, which covers the most common use cases:
lambdas and method groups:
```
// Lambda
Callable.From((int num) => "Foo" + num);
// Method group
string AppendNum(int num) => "Foo" + num;
Callable.From(AppendNum);
```
Unfortunately, due to limitations in the C# language, implicit
conversions from delegates to `Callable` are not supported.
`Callable.From` does not support custom delegates. These should be
uncommon, but the Godot C# API actually uses them for event signals.
As such, the bindings generator was updated to generate trampoline
functions for event signals. It was also optimized to use `Action`
instead of a custom delegate for parameterless signals, which removes
the need for the trampoline functions for those signals.
The change to reflection-free invokation removes one of the last needs
for `ConvertVariantToManagedObjectOfType`. The only remaining usage is
from calling script constructors with parameters from the engine
(`CreateManagedForGodotObjectScriptInstance`). Once that one is made
reflection-free, `ConvertVariantToManagedObjectOfType` can be removed.
2022-10-29 04:59:13 +08:00
< < " ) && args.Count == " < < itos ( imethod . arguments . size ( ) )
2021-12-29 06:25:16 +08:00
< < " && " < < CS_METHOD_HAS_GODOT_CLASS_METHOD < < " ((godot_string_name) "
< < CS_STATIC_FIELD_METHOD_PROXY_NAME_PREFIX < < imethod . name < < " .NativeValue)) \n "
2022-02-28 04:57:50 +08:00
< < INDENT2 " { \n " ;
2021-09-13 02:21:15 +08:00
if ( imethod . return_type . cname ! = name_cache . type_void ) {
2022-02-28 04:57:50 +08:00
output < < INDENT3 " var callRet = " ;
2021-09-13 02:21:15 +08:00
} else {
2022-02-28 04:57:50 +08:00
output < < INDENT3 ;
2021-09-13 02:21:15 +08:00
}
output < < imethod . proxy_name < < " ( " ;
for ( int i = 0 ; i < imethod . arguments . size ( ) ; i + + ) {
const ArgumentInterface & iarg = imethod . arguments [ i ] ;
const TypeInterface * arg_type = _get_type_or_null ( iarg . type ) ;
ERR_FAIL_NULL_V ( arg_type , ERR_BUG ) ; // Argument type not found
if ( i ! = 0 ) {
output < < " , " ;
}
if ( arg_type - > cname = = name_cache . type_Array_generic | | arg_type - > cname = = name_cache . type_Dictionary_generic ) {
String arg_cs_type = arg_type - > cs_type + _get_generic_type_parameters ( * arg_type , iarg . type . generic_type_parameters ) ;
2022-07-28 23:41:50 +08:00
output < < " new " < < arg_cs_type < < " ( " < < sformat ( arg_type - > cs_variant_to_managed , " args[ " + itos ( i ) + " ] " , arg_type - > cs_type , arg_type - > name ) < < " ) " ;
2021-09-13 02:21:15 +08:00
} else {
2022-02-28 04:57:46 +08:00
output < < sformat ( arg_type - > cs_variant_to_managed ,
" args[ " + itos ( i ) + " ] " , arg_type - > cs_type , arg_type - > name ) ;
2021-09-13 02:21:15 +08:00
}
}
output < < " ); \n " ;
if ( imethod . return_type . cname ! = name_cache . type_void ) {
2022-02-28 04:57:46 +08:00
const TypeInterface * return_type = _get_type_or_null ( imethod . return_type ) ;
ERR_FAIL_NULL_V ( return_type , ERR_BUG ) ; // Return type not found
2022-02-28 04:57:50 +08:00
output < < INDENT3 " ret = "
2022-02-28 04:57:46 +08:00
< < sformat ( return_type - > cs_managed_to_variant , " callRet " , return_type - > cs_type , return_type - > name )
< < " ; \n "
2022-02-28 04:57:50 +08:00
< < INDENT3 " return true; \n " ;
2021-09-13 02:21:15 +08:00
} else {
2022-02-28 04:57:50 +08:00
output < < INDENT3 " ret = default; \n "
< < INDENT3 " return true; \n " ;
2021-09-13 02:21:15 +08:00
}
2022-02-28 04:57:50 +08:00
output < < INDENT2 " } \n " ;
2021-09-13 02:21:15 +08:00
}
if ( is_derived_type ) {
C#: Remove need for reflection to invoking callable delegates
We aim to make the C# API reflection-free, mainly for concerns about
performance, and to be able to target NativeAOT in refletion-free mode,
which reduces the binary size.
One of the main usages of reflection still left was the dynamic
invokation of callable delegates, and for some time I wasn't sure
I would find an alternative solution that I'd be happy with.
The new solution uses trampoline functions to invoke the delegates:
```
static void Trampoline(object delegateObj, NativeVariantPtrArgs args, out godot_variant ret)
{
if (args.Count != 1)
throw new ArgumentException($"Callable expected 1 arguments but received {args.Count}.");
string res = ((Func<int, string>)delegateObj)(
VariantConversionCallbacks.GetToManagedCallback<int>()(args[0])
);
ret = VariantConversionCallbacks.GetToVariantCallback<string>()(res);
}
Callable.CreateWithUnsafeTrampoline((int num) => "Foo" + num, &Trampoline);
```
Of course, this is too much boilerplate for user code. To improve this,
the `Callable.From` methods were added. These are overloads that take
`Action` and `Func` delegates, which covers the most common use cases:
lambdas and method groups:
```
// Lambda
Callable.From((int num) => "Foo" + num);
// Method group
string AppendNum(int num) => "Foo" + num;
Callable.From(AppendNum);
```
Unfortunately, due to limitations in the C# language, implicit
conversions from delegates to `Callable` are not supported.
`Callable.From` does not support custom delegates. These should be
uncommon, but the Godot C# API actually uses them for event signals.
As such, the bindings generator was updated to generate trampoline
functions for event signals. It was also optimized to use `Action`
instead of a custom delegate for parameterless signals, which removes
the need for the trampoline functions for those signals.
The change to reflection-free invokation removes one of the last needs
for `ConvertVariantToManagedObjectOfType`. The only remaining usage is
from calling script constructors with parameters from the engine
(`CreateManagedForGodotObjectScriptInstance`). Once that one is made
reflection-free, `ConvertVariantToManagedObjectOfType` can be removed.
2022-10-29 04:59:13 +08:00
output < < INDENT2 " return base. " CS_METHOD_INVOKE_GODOT_CLASS_METHOD " (method, args, out ret); \n " ;
2021-09-13 02:21:15 +08:00
} else {
2022-02-28 04:57:50 +08:00
output < < INDENT2 " ret = default; \n "
< < INDENT2 " return false; \n " ;
2021-09-13 02:21:15 +08:00
}
2022-02-28 04:57:50 +08:00
output < < INDENT1 " } \n " ;
2021-12-29 06:25:16 +08:00
// Generate HasGodotClassMethod
output < < MEMBER_BEGIN " protected internal " < < ( is_derived_type ? " override " : " virtual " )
< < " bool " CS_METHOD_HAS_GODOT_CLASS_METHOD " (in godot_string_name method) \n "
2022-02-28 04:57:50 +08:00
< < INDENT1 " { \n " ;
2021-12-29 06:25:16 +08:00
for ( const MethodInterface & imethod : itype . methods ) {
if ( ! imethod . is_virtual ) {
continue ;
}
// We check for native names (snake_case). If we detect one, we call HasGodotClassMethod
// again, but this time with the respective proxy name (PascalCase). It's the job of
// user derived classes to override the method and check for those. Our C# source
// generators take care of generating those override methods.
2022-08-25 23:07:20 +08:00
output < < INDENT2 " if (method == MethodName. " < < imethod . proxy_name
2022-02-28 04:57:50 +08:00
< < " ) \n " INDENT2 " { \n "
< < INDENT3 " if ( " CS_METHOD_HAS_GODOT_CLASS_METHOD " ( "
2021-12-29 06:25:16 +08:00
< < CS_STATIC_FIELD_METHOD_PROXY_NAME_PREFIX < < imethod . name
2022-02-28 04:57:50 +08:00
< < " .NativeValue.DangerousSelfRef)) \n " INDENT3 " { \n "
< < INDENT4 " return true; \n "
< < INDENT3 " } \n " INDENT2 " } \n " ;
2021-12-29 06:25:16 +08:00
}
if ( is_derived_type ) {
2022-02-28 04:57:50 +08:00
output < < INDENT2 " return base. " CS_METHOD_HAS_GODOT_CLASS_METHOD " (method); \n " ;
2021-12-29 06:25:16 +08:00
} else {
2022-02-28 04:57:50 +08:00
output < < INDENT2 " return false; \n " ;
2021-12-29 06:25:16 +08:00
}
2022-02-28 04:57:50 +08:00
output < < INDENT1 " } \n " ;
2023-01-14 03:59:02 +08:00
// Generate HasGodotClassSignal
output < < MEMBER_BEGIN " protected internal " < < ( is_derived_type ? " override " : " virtual " )
< < " bool " CS_METHOD_HAS_GODOT_CLASS_SIGNAL " (in godot_string_name signal) \n "
< < INDENT1 " { \n " ;
for ( const SignalInterface & isignal : itype . signals_ ) {
// We check for native names (snake_case). If we detect one, we call HasGodotClassSignal
// again, but this time with the respective proxy name (PascalCase). It's the job of
// user derived classes to override the method and check for those. Our C# source
// generators take care of generating those override methods.
output < < INDENT2 " if (signal == SignalName. " < < isignal . proxy_name
< < " ) \n " INDENT2 " { \n "
< < INDENT3 " if ( " CS_METHOD_HAS_GODOT_CLASS_SIGNAL " ( "
< < CS_STATIC_FIELD_SIGNAL_PROXY_NAME_PREFIX < < isignal . name
< < " .NativeValue.DangerousSelfRef)) \n " INDENT3 " { \n "
< < INDENT4 " return true; \n "
< < INDENT3 " } \n " INDENT2 " } \n " ;
}
if ( is_derived_type ) {
output < < INDENT2 " return base. " CS_METHOD_HAS_GODOT_CLASS_SIGNAL " (signal); \n " ;
} else {
output < < INDENT2 " return false; \n " ;
}
output < < INDENT1 " } \n " ;
2021-09-13 02:21:15 +08:00
}
2022-08-25 23:07:20 +08:00
//Generate StringName for all class members
bool is_inherit = ! itype . is_singleton & & obj_types . has ( itype . base_name ) ;
//PropertyName
if ( is_inherit ) {
output < < MEMBER_BEGIN " public new class PropertyName : " < < obj_types [ itype . base_name ] . proxy_name < < " .PropertyName " ;
} else {
output < < MEMBER_BEGIN " public class PropertyName " ;
}
output < < " \n "
< < INDENT1 " { \n " ;
for ( const PropertyInterface & iprop : itype . properties ) {
output < < INDENT2 " public static readonly StringName " < < iprop . proxy_name < < " = \" " < < iprop . cname < < " \" ; \n " ;
}
output < < INDENT1 " } \n " ;
//MethodName
if ( is_inherit ) {
output < < MEMBER_BEGIN " public new class MethodName : " < < obj_types [ itype . base_name ] . proxy_name < < " .MethodName " ;
} else {
output < < MEMBER_BEGIN " public class MethodName " ;
}
output < < " \n "
< < INDENT1 " { \n " ;
for ( const MethodInterface & imethod : itype . methods ) {
output < < INDENT2 " public static readonly StringName " < < imethod . proxy_name < < " = \" " < < imethod . cname < < " \" ; \n " ;
}
output < < INDENT1 " } \n " ;
//SignalName
if ( is_inherit ) {
output < < MEMBER_BEGIN " public new class SignalName : " < < obj_types [ itype . base_name ] . proxy_name < < " .SignalName " ;
} else {
output < < MEMBER_BEGIN " public class SignalName " ;
}
output < < " \n "
< < INDENT1 " { \n " ;
for ( const SignalInterface & isignal : itype . signals_ ) {
output < < INDENT2 " public static readonly StringName " < < isignal . proxy_name < < " = \" " < < isignal . cname < < " \" ; \n " ;
}
output < < INDENT1 " } \n " ;
2022-02-28 04:57:50 +08:00
output . append ( CLOSE_BLOCK /* class */ ) ;
2018-08-28 02:39:51 +08:00
2019-04-19 07:10:08 +08:00
output . append ( " \n "
" #pragma warning restore CS1591 \n "
" #pragma warning restore CS1573 \n " ) ;
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
return _save_file ( p_output_file , output ) ;
}
2017-10-03 05:24:00 +08:00
2019-04-19 07:10:08 +08:00
Error BindingsGenerator : : _generate_cs_property ( const BindingsGenerator : : TypeInterface & p_itype , const PropertyInterface & p_iprop , StringBuilder & p_output ) {
2017-12-24 10:17:48 +08:00
const MethodInterface * setter = p_itype . find_method_by_name ( p_iprop . setter ) ;
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
// Search it in base types too
const TypeInterface * current_type = & p_itype ;
2017-12-24 10:17:48 +08:00
while ( ! setter & & current_type - > base_name ! = StringName ( ) ) {
2022-05-08 16:09:19 +08:00
HashMap < StringName , TypeInterface > : : Iterator base_match = obj_types . find ( current_type - > base_name ) ;
2020-04-03 07:38:48 +08:00
ERR_FAIL_COND_V_MSG ( ! base_match , ERR_BUG , " Type not found ' " + current_type - > base_name + " '. Inherited by ' " + current_type - > name + " '. " ) ;
2022-05-08 16:09:19 +08:00
current_type = & base_match - > value ;
2017-12-24 10:17:48 +08:00
setter = current_type - > find_method_by_name ( p_iprop . setter ) ;
2017-10-29 09:37:13 +08:00
}
2017-10-03 05:24:00 +08:00
2017-12-24 10:17:48 +08:00
const MethodInterface * getter = p_itype . find_method_by_name ( p_iprop . getter ) ;
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
// Search it in base types too
current_type = & p_itype ;
2017-12-24 10:17:48 +08:00
while ( ! getter & & current_type - > base_name ! = StringName ( ) ) {
2022-05-08 16:09:19 +08:00
HashMap < StringName , TypeInterface > : : Iterator base_match = obj_types . find ( current_type - > base_name ) ;
2020-04-03 07:38:48 +08:00
ERR_FAIL_COND_V_MSG ( ! base_match , ERR_BUG , " Type not found ' " + current_type - > base_name + " '. Inherited by ' " + current_type - > name + " '. " ) ;
2022-05-08 16:09:19 +08:00
current_type = & base_match - > value ;
2017-12-24 10:17:48 +08:00
getter = current_type - > find_method_by_name ( p_iprop . getter ) ;
2017-10-29 09:37:13 +08:00
}
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
ERR_FAIL_COND_V ( ! setter & & ! getter , ERR_BUG ) ;
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
if ( setter ) {
2017-12-24 10:17:48 +08:00
int setter_argc = p_iprop . index ! = - 1 ? 2 : 1 ;
2017-10-29 09:37:13 +08:00
ERR_FAIL_COND_V ( setter - > arguments . size ( ) ! = setter_argc , ERR_BUG ) ;
}
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
if ( getter ) {
2017-12-24 10:17:48 +08:00
int getter_argc = p_iprop . index ! = - 1 ? 1 : 0 ;
2017-10-29 09:37:13 +08:00
ERR_FAIL_COND_V ( getter - > arguments . size ( ) ! = getter_argc , ERR_BUG ) ;
}
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
if ( getter & & setter ) {
2020-03-15 02:20:17 +08:00
const ArgumentInterface & setter_first_arg = setter - > arguments . back ( ) - > get ( ) ;
if ( getter - > return_type . cname ! = setter_first_arg . type . cname ) {
// Special case for Node::set_name
bool whitelisted = getter - > return_type . cname = = name_cache . type_StringName & &
2021-10-28 21:19:35 +08:00
setter_first_arg . type . cname = = name_cache . type_String ;
2020-03-15 02:20:17 +08:00
ERR_FAIL_COND_V_MSG ( ! whitelisted , ERR_BUG ,
" Return type from getter doesn't match first argument of setter for property: ' " +
p_itype . name + " . " + String ( p_iprop . cname ) + " '. " ) ;
}
2017-10-29 09:37:13 +08:00
}
2017-10-03 05:24:00 +08:00
2018-04-29 04:25:25 +08:00
const TypeReference & proptype_name = getter ? getter - > return_type : setter - > arguments . back ( ) - > get ( ) . type ;
2017-10-03 05:24:00 +08:00
2018-04-29 04:25:25 +08:00
const TypeInterface * prop_itype = _get_type_or_null ( proptype_name ) ;
2017-12-24 10:17:48 +08:00
ERR_FAIL_NULL_V ( prop_itype , ERR_BUG ) ; // Property type not found
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
ERR_FAIL_COND_V_MSG ( prop_itype - > is_singleton , ERR_BUG ,
" Property type is a singleton: ' " + p_itype . name + " . " + String ( p_iprop . cname ) + " '. " ) ;
2021-01-09 07:14:25 +08:00
if ( p_itype . api_type = = ClassDB : : API_CORE ) {
ERR_FAIL_COND_V_MSG ( prop_itype - > api_type = = ClassDB : : API_EDITOR , ERR_BUG ,
" Property ' " + p_itype . name + " . " + String ( p_iprop . cname ) + " ' has type ' " + prop_itype - > name +
" ' from the editor API. Core API cannot have dependencies on the editor API. " ) ;
}
2017-12-24 10:17:48 +08:00
if ( p_iprop . prop_doc & & p_iprop . prop_doc - > description . size ( ) ) {
2019-03-19 04:59:35 +08:00
String xml_summary = bbcode_to_xml ( fix_doc_description ( p_iprop . prop_doc - > description ) , & p_itype ) ;
2019-04-19 07:10:08 +08:00
Vector < String > summary_lines = xml_summary . length ( ) ? xml_summary . split ( " \n " ) : Vector < String > ( ) ;
2017-10-03 05:24:00 +08:00
2019-03-19 04:59:35 +08:00
if ( summary_lines . size ( ) ) {
2019-04-19 07:10:08 +08:00
p_output . append ( MEMBER_BEGIN " /// <summary> \n " ) ;
2017-10-03 05:24:00 +08:00
2019-03-19 04:59:35 +08:00
for ( int i = 0 ; i < summary_lines . size ( ) ; i + + ) {
2022-02-28 04:57:50 +08:00
p_output . append ( INDENT1 " /// " ) ;
2019-04-19 07:10:08 +08:00
p_output . append ( summary_lines [ i ] ) ;
p_output . append ( " \n " ) ;
2017-10-29 09:37:13 +08:00
}
2017-10-03 05:24:00 +08:00
2022-02-28 04:57:50 +08:00
p_output . append ( INDENT1 " /// </summary> " ) ;
2019-03-19 04:59:35 +08:00
}
2017-10-29 09:37:13 +08:00
}
2017-10-03 05:24:00 +08:00
2019-04-19 07:10:08 +08:00
p_output . append ( MEMBER_BEGIN " public " ) ;
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
if ( p_itype . is_singleton ) {
2019-04-19 07:10:08 +08:00
p_output . append ( " static " ) ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
2022-02-09 07:56:10 +08:00
String prop_cs_type = prop_itype - > cs_type + _get_generic_type_parameters ( * prop_itype , proptype_name . generic_type_parameters ) ;
p_output . append ( prop_cs_type ) ;
2019-04-19 07:10:08 +08:00
p_output . append ( " " ) ;
p_output . append ( p_iprop . proxy_name ) ;
2022-02-28 04:57:50 +08:00
p_output . append ( " \n " OPEN_BLOCK_L1 ) ;
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
if ( getter ) {
2023-01-16 21:57:50 +08:00
p_output . append ( INDENT2 " get \n " OPEN_BLOCK_L2 INDENT3 ) ;
2019-04-25 23:11:01 +08:00
2019-04-19 07:10:08 +08:00
p_output . append ( " return " ) ;
p_output . append ( getter - > proxy_name + " ( " ) ;
2017-12-24 10:17:48 +08:00
if ( p_iprop . index ! = - 1 ) {
const ArgumentInterface & idx_arg = getter - > arguments . front ( ) - > get ( ) ;
2018-04-29 04:25:25 +08:00
if ( idx_arg . type . cname ! = name_cache . type_int ) {
2017-12-24 10:17:48 +08:00
// Assume the index parameter is an enum
2018-04-29 04:25:25 +08:00
const TypeInterface * idx_arg_type = _get_type_or_null ( idx_arg . type ) ;
2020-04-02 07:20:12 +08:00
CRASH_COND ( idx_arg_type = = nullptr ) ;
2019-04-19 07:10:08 +08:00
p_output . append ( " ( " + idx_arg_type - > proxy_name + " ) " + itos ( p_iprop . index ) ) ;
2017-12-24 10:17:48 +08:00
} else {
2019-04-19 07:10:08 +08:00
p_output . append ( itos ( p_iprop . index ) ) ;
2017-12-24 10:17:48 +08:00
}
}
2023-01-16 21:57:50 +08:00
p_output . append ( " ); \n " CLOSE_BLOCK_L2 ) ;
2017-10-29 09:37:13 +08:00
}
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
if ( setter ) {
2023-01-16 21:57:50 +08:00
p_output . append ( INDENT2 " set \n " OPEN_BLOCK_L2 INDENT3 ) ;
2019-04-25 23:11:01 +08:00
2019-04-19 07:10:08 +08:00
p_output . append ( setter - > proxy_name + " ( " ) ;
2017-12-24 10:17:48 +08:00
if ( p_iprop . index ! = - 1 ) {
const ArgumentInterface & idx_arg = setter - > arguments . front ( ) - > get ( ) ;
2018-04-29 04:25:25 +08:00
if ( idx_arg . type . cname ! = name_cache . type_int ) {
2017-12-24 10:17:48 +08:00
// Assume the index parameter is an enum
2018-04-29 04:25:25 +08:00
const TypeInterface * idx_arg_type = _get_type_or_null ( idx_arg . type ) ;
2020-04-02 07:20:12 +08:00
CRASH_COND ( idx_arg_type = = nullptr ) ;
2019-04-19 07:10:08 +08:00
p_output . append ( " ( " + idx_arg_type - > proxy_name + " ) " + itos ( p_iprop . index ) + " , " ) ;
2017-12-24 10:17:48 +08:00
} else {
2019-04-19 07:10:08 +08:00
p_output . append ( itos ( p_iprop . index ) + " , " ) ;
2017-12-24 10:17:48 +08:00
}
}
2023-01-16 21:57:50 +08:00
p_output . append ( " value); \n " CLOSE_BLOCK_L2 ) ;
2017-10-29 09:37:13 +08:00
}
2017-10-03 05:24:00 +08:00
2022-02-28 04:57:50 +08:00
p_output . append ( CLOSE_BLOCK_L1 ) ;
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
return OK ;
}
2017-10-03 05:24:00 +08:00
2019-04-19 07:10:08 +08:00
Error BindingsGenerator : : _generate_cs_method ( const BindingsGenerator : : TypeInterface & p_itype , const BindingsGenerator : : MethodInterface & p_imethod , int & p_method_bind_count , StringBuilder & p_output ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
const TypeInterface * return_type = _get_type_or_null ( p_imethod . return_type ) ;
ERR_FAIL_NULL_V ( return_type , ERR_BUG ) ; // Return type not found
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
ERR_FAIL_COND_V_MSG ( return_type - > is_singleton , ERR_BUG ,
" Method return type is a singleton: ' " + p_itype . name + " . " + p_imethod . name + " '. " ) ;
2021-01-09 07:14:25 +08:00
if ( p_itype . api_type = = ClassDB : : API_CORE ) {
ERR_FAIL_COND_V_MSG ( return_type - > api_type = = ClassDB : : API_EDITOR , ERR_BUG ,
" Method ' " + p_itype . name + " . " + p_imethod . name + " ' has return type ' " + return_type - > name +
" ' from the editor API. Core API cannot have dependencies on the editor API. " ) ;
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
String method_bind_field = CS_STATIC_FIELD_METHOD_BIND_PREFIX + itos ( p_method_bind_count ) ;
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
String arguments_sig ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
StringBuilder cs_in_statements ;
2021-12-29 06:25:16 +08:00
bool cs_in_expr_is_unsafe = false ;
2017-10-03 05:24:00 +08:00
2022-03-21 08:25:50 +08:00
String icall_params = method_bind_field ;
2021-12-29 06:25:16 +08:00
2022-03-21 08:25:50 +08:00
if ( ! p_imethod . is_static ) {
2021-12-29 06:25:16 +08:00
if ( p_itype . cs_in . size ( ) ) {
cs_in_statements < < sformat ( p_itype . cs_in , p_itype . c_type , " this " ,
2022-02-28 04:57:50 +08:00
String ( ) , String ( ) , String ( ) , INDENT2 ) ;
2021-12-29 06:25:16 +08:00
}
icall_params + = " , " + sformat ( p_itype . cs_in_expr , " this " ) ;
2022-03-21 08:25:50 +08:00
}
2018-01-26 06:44:37 +08:00
2019-04-19 07:10:08 +08:00
StringBuilder default_args_doc ;
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
// Retrieve information from the arguments
2021-07-26 00:45:00 +08:00
const ArgumentInterface & first = p_imethod . arguments . front ( ) - > get ( ) ;
for ( const ArgumentInterface & iarg : p_imethod . arguments ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
const TypeInterface * arg_type = _get_type_or_null ( iarg . type ) ;
ERR_FAIL_NULL_V ( arg_type , ERR_BUG ) ; // Argument type not found
2017-10-29 09:37:13 +08:00
2020-05-16 10:03:05 +08:00
ERR_FAIL_COND_V_MSG ( arg_type - > is_singleton , ERR_BUG ,
" Argument type is a singleton: ' " + iarg . name + " ' of method ' " + p_itype . name + " . " + p_imethod . name + " '. " ) ;
2021-01-09 07:14:25 +08:00
if ( p_itype . api_type = = ClassDB : : API_CORE ) {
ERR_FAIL_COND_V_MSG ( arg_type - > api_type = = ClassDB : : API_EDITOR , ERR_BUG ,
" Argument ' " + iarg . name + " ' of method ' " + p_itype . name + " . " + p_imethod . name + " ' has type ' " +
arg_type - > name + " ' from the editor API. Core API cannot have dependencies on the editor API. " ) ;
}
2020-05-16 10:03:05 +08:00
if ( iarg . default_argument . size ( ) ) {
CRASH_COND_MSG ( ! _arg_default_value_is_assignable_to_type ( iarg . def_param_value , * arg_type ) ,
" Invalid default value for parameter ' " + iarg . name + " ' of method ' " + p_itype . name + " . " + p_imethod . name + " '. " ) ;
}
2022-02-09 07:56:10 +08:00
String arg_cs_type = arg_type - > cs_type + _get_generic_type_parameters ( * arg_type , iarg . type . generic_type_parameters ) ;
2017-10-29 09:37:13 +08:00
// Add the current arguments to the signature
// If the argument has a default value which is not a constant, we will make it Nullable
{
2021-07-26 00:45:00 +08:00
if ( & iarg ! = & first ) {
2017-10-29 09:37:13 +08:00
arguments_sig + = " , " ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
if ( iarg . def_param_mode = = ArgumentInterface : : NULLABLE_VAL ) {
2017-10-29 09:37:13 +08:00
arguments_sig + = " Nullable< " ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
2022-02-09 07:56:10 +08:00
arguments_sig + = arg_cs_type ;
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
if ( iarg . def_param_mode = = ArgumentInterface : : NULLABLE_VAL ) {
2017-10-29 09:37:13 +08:00
arguments_sig + = " > " ;
2020-05-16 10:03:05 +08:00
} else {
2017-10-29 09:37:13 +08:00
arguments_sig + = " " ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
arguments_sig + = iarg . name ;
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
if ( iarg . default_argument . size ( ) ) {
2020-05-16 10:03:05 +08:00
if ( iarg . def_param_mode ! = ArgumentInterface : : CONSTANT ) {
2017-10-29 09:37:13 +08:00
arguments_sig + = " = null " ;
2020-05-16 10:03:05 +08:00
} else {
2017-10-29 09:37:13 +08:00
arguments_sig + = " = " + sformat ( iarg . default_argument , arg_type - > cs_type ) ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
}
}
2017-10-29 09:37:13 +08:00
icall_params + = " , " ;
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
if ( iarg . default_argument . size ( ) & & iarg . def_param_mode ! = ArgumentInterface : : CONSTANT ) {
// The default value of an argument must be constant. Otherwise we make it Nullable and do the following:
// Type arg_in = arg.HasValue ? arg.Value : <non-const default value>;
2021-12-29 06:25:16 +08:00
String arg_or_defval_local = iarg . name ;
arg_or_defval_local + = " OrDefVal " ;
2017-10-03 05:24:00 +08:00
2022-02-28 04:57:50 +08:00
cs_in_statements < < INDENT2 < < arg_cs_type < < " " < < arg_or_defval_local < < " = " < < iarg . name ;
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
if ( iarg . def_param_mode = = ArgumentInterface : : NULLABLE_VAL ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
cs_in_statements < < " .HasValue ? " ;
2020-05-16 10:03:05 +08:00
} else {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
cs_in_statements < < " != null ? " ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
cs_in_statements < < iarg . name ;
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
if ( iarg . def_param_mode = = ArgumentInterface : : NULLABLE_VAL ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
cs_in_statements < < " .Value : " ;
2020-05-16 10:03:05 +08:00
} else {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
cs_in_statements < < " : " ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
2022-02-09 07:56:10 +08:00
String cs_type = arg_cs_type ;
2021-07-25 19:46:31 +08:00
if ( cs_type . ends_with ( " [] " ) ) {
cs_type = cs_type . substr ( 0 , cs_type . length ( ) - 2 ) ;
}
String def_arg = sformat ( iarg . default_argument , cs_type ) ;
2017-10-03 05:24:00 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
cs_in_statements < < def_arg < < " ; \n " ;
2017-10-03 05:24:00 +08:00
2021-12-29 06:25:16 +08:00
if ( arg_type - > cs_in . size ( ) ) {
cs_in_statements < < sformat ( arg_type - > cs_in , arg_type - > c_type , arg_or_defval_local ,
2022-02-28 04:57:50 +08:00
String ( ) , String ( ) , String ( ) , INDENT2 ) ;
2021-12-29 06:25:16 +08:00
}
if ( arg_type - > cs_in_expr . is_empty ( ) ) {
icall_params + = arg_or_defval_local ;
} else {
icall_params + = sformat ( arg_type - > cs_in_expr , arg_or_defval_local , arg_type - > c_type ) ;
}
2017-10-03 05:24:00 +08:00
2019-03-19 04:59:35 +08:00
// Apparently the name attribute must not include the @
String param_tag_name = iarg . name . begins_with ( " @ " ) ? iarg . name . substr ( 1 , iarg . name . length ( ) ) : iarg . name ;
2021-07-25 19:46:31 +08:00
// Escape < and > in the attribute default value
String param_def_arg = def_arg . replacen ( " < " , " < " ) . replacen ( " > " , " > " ) ;
2019-03-19 04:59:35 +08:00
2022-02-09 07:56:10 +08:00
default_args_doc . append ( MEMBER_BEGIN " /// <param name= \" " + param_tag_name + " \" >If the parameter is null, then the default value is <c> " + param_def_arg + " </c>.</param> " ) ;
2017-10-29 09:37:13 +08:00
} else {
2021-12-29 06:25:16 +08:00
if ( arg_type - > cs_in . size ( ) ) {
cs_in_statements < < sformat ( arg_type - > cs_in , arg_type - > c_type , iarg . name ,
2022-02-28 04:57:50 +08:00
String ( ) , String ( ) , String ( ) , INDENT2 ) ;
2021-12-29 06:25:16 +08:00
}
icall_params + = arg_type - > cs_in_expr . is_empty ( ) ? iarg . name : sformat ( arg_type - > cs_in_expr , iarg . name , arg_type - > c_type ) ;
2017-10-29 09:37:13 +08:00
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
2021-12-29 06:25:16 +08:00
cs_in_expr_is_unsafe | = arg_type - > cs_in_expr_is_unsafe ;
2017-10-29 09:37:13 +08:00
}
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
// Generate method
{
2018-09-04 11:40:41 +08:00
if ( ! p_imethod . is_virtual & & ! p_imethod . requires_object_call ) {
2021-12-29 06:25:16 +08:00
p_output < < MEMBER_BEGIN " [DebuggerBrowsable(DebuggerBrowsableState.Never)] \n "
2022-02-28 04:57:50 +08:00
< < INDENT1 " private static readonly IntPtr " < < method_bind_field < < " = " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
if ( p_itype . is_singleton ) {
2022-12-07 23:16:51 +08:00
// Singletons are static classes. They don't derive GodotObject,
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
// so we need to specify the type to call the static method.
2022-12-07 23:16:51 +08:00
p_output < < " GodotObject. " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
}
2022-08-25 23:07:20 +08:00
p_output < < ICALL_CLASSDB_GET_METHOD " ( " BINDINGS_NATIVE_NAME_FIELD " , MethodName. "
< < p_imethod . proxy_name
< < " ); \n " ;
2017-10-29 09:37:13 +08:00
}
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
if ( p_imethod . method_doc & & p_imethod . method_doc - > description . size ( ) ) {
2019-03-19 04:59:35 +08:00
String xml_summary = bbcode_to_xml ( fix_doc_description ( p_imethod . method_doc - > description ) , & p_itype ) ;
2019-04-19 07:10:08 +08:00
Vector < String > summary_lines = xml_summary . length ( ) ? xml_summary . split ( " \n " ) : Vector < String > ( ) ;
2017-10-03 05:24:00 +08:00
2020-01-03 01:09:59 +08:00
if ( summary_lines . size ( ) ) {
2019-04-19 07:10:08 +08:00
p_output . append ( MEMBER_BEGIN " /// <summary> \n " ) ;
2017-10-03 05:24:00 +08:00
2019-03-19 04:59:35 +08:00
for ( int i = 0 ; i < summary_lines . size ( ) ; i + + ) {
2022-02-28 04:57:50 +08:00
p_output . append ( INDENT1 " /// " ) ;
2019-04-19 07:10:08 +08:00
p_output . append ( summary_lines [ i ] ) ;
p_output . append ( " \n " ) ;
2017-10-29 09:37:13 +08:00
}
2017-10-03 05:24:00 +08:00
2022-02-28 04:57:50 +08:00
p_output . append ( INDENT1 " /// </summary> " ) ;
2019-03-19 04:59:35 +08:00
}
2017-10-29 09:37:13 +08:00
}
2017-10-03 05:24:00 +08:00
2020-01-03 01:09:59 +08:00
if ( default_args_doc . get_string_length ( ) ) {
p_output . append ( default_args_doc . as_string ( ) ) ;
}
2019-04-25 23:11:01 +08:00
if ( p_imethod . is_deprecated ) {
2020-12-15 20:04:21 +08:00
if ( p_imethod . deprecation_message . is_empty ( ) ) {
2019-11-07 16:44:15 +08:00
WARN_PRINT ( " An empty deprecation message is discouraged. Method: ' " + p_imethod . proxy_name + " '. " ) ;
2020-05-16 10:03:05 +08:00
}
2019-04-25 23:11:01 +08:00
p_output . append ( MEMBER_BEGIN " [Obsolete( \" " ) ;
p_output . append ( p_imethod . deprecation_message ) ;
p_output . append ( " \" )] " ) ;
}
2019-04-19 07:10:08 +08:00
p_output . append ( MEMBER_BEGIN ) ;
p_output . append ( p_imethod . is_internal ? " internal " : " public " ) ;
2017-10-03 05:24:00 +08:00
2022-03-21 08:25:50 +08:00
if ( p_itype . is_singleton | | p_imethod . is_static ) {
2019-04-19 07:10:08 +08:00
p_output . append ( " static " ) ;
2017-10-29 09:37:13 +08:00
} else if ( p_imethod . is_virtual ) {
2019-04-19 07:10:08 +08:00
p_output . append ( " virtual " ) ;
2017-10-29 09:37:13 +08:00
}
2017-10-03 05:24:00 +08:00
2021-12-29 06:25:16 +08:00
if ( cs_in_expr_is_unsafe ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
p_output . append ( " unsafe " ) ;
}
2022-02-09 07:56:10 +08:00
String return_cs_type = return_type - > cs_type + _get_generic_type_parameters ( * return_type , p_imethod . return_type . generic_type_parameters ) ;
p_output . append ( return_cs_type + " " ) ;
2019-04-19 07:10:08 +08:00
p_output . append ( p_imethod . proxy_name + " ( " ) ;
2022-02-28 04:57:50 +08:00
p_output . append ( arguments_sig + " ) \n " OPEN_BLOCK_L1 ) ;
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
if ( p_imethod . is_virtual ) {
// Godot virtual method must be overridden, therefore we return a default value by default.
2017-10-03 05:24:00 +08:00
2017-12-24 10:17:48 +08:00
if ( return_type - > cname = = name_cache . type_void ) {
2022-02-28 04:57:50 +08:00
p_output . append ( CLOSE_BLOCK_L1 ) ;
2017-10-29 09:37:13 +08:00
} else {
2022-02-28 04:57:50 +08:00
p_output . append ( INDENT2 " return default; \n " CLOSE_BLOCK_L1 ) ;
2017-10-29 09:37:13 +08:00
}
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
return OK ; // Won't increment method bind count
}
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
if ( p_imethod . requires_object_call ) {
// Fallback to Godot's object.Call(string, params)
2017-10-03 05:24:00 +08:00
2022-02-28 04:57:50 +08:00
p_output . append ( INDENT2 CS_METHOD_CALL " ( \" " ) ;
2019-04-19 07:10:08 +08:00
p_output . append ( p_imethod . name ) ;
p_output . append ( " \" " ) ;
2017-10-03 05:24:00 +08:00
2021-07-26 00:45:00 +08:00
for ( const ArgumentInterface & iarg : p_imethod . arguments ) {
2019-04-19 07:10:08 +08:00
p_output . append ( " , " ) ;
2021-07-26 00:45:00 +08:00
p_output . append ( iarg . name ) ;
2017-10-29 09:37:13 +08:00
}
2017-10-03 05:24:00 +08:00
2022-02-28 04:57:50 +08:00
p_output . append ( " ); \n " CLOSE_BLOCK_L1 ) ;
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
return OK ; // Won't increment method bind count
}
2017-10-03 05:24:00 +08:00
2022-05-13 21:04:37 +08:00
HashMap < const MethodInterface * , const InternalCall * > : : ConstIterator match = method_icalls_map . find ( & p_imethod ) ;
2017-10-29 09:37:13 +08:00
ERR_FAIL_NULL_V ( match , ERR_BUG ) ;
2017-10-03 05:24:00 +08:00
2022-05-13 21:04:37 +08:00
const InternalCall * im_icall = match - > value ;
2017-10-03 05:24:00 +08:00
2018-02-22 20:13:51 +08:00
String im_call = im_icall - > editor_only ? BINDINGS_CLASS_NATIVECALLS_EDITOR : BINDINGS_CLASS_NATIVECALLS ;
2019-11-15 04:48:49 +08:00
im_call + = " . " ;
im_call + = im_icall - > name ;
2017-10-03 05:24:00 +08:00
2021-12-29 06:25:16 +08:00
if ( p_imethod . arguments . size ( ) & & cs_in_statements . get_string_length ( ) > 0 ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
p_output . append ( cs_in_statements . as_string ( ) ) ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
2017-12-24 10:17:48 +08:00
if ( return_type - > cname = = name_cache . type_void ) {
2022-02-28 04:57:50 +08:00
p_output < < INDENT2 < < im_call < < " ( " < < icall_params < < " ); \n " ;
2020-12-15 20:04:21 +08:00
} else if ( return_type - > cs_out . is_empty ( ) ) {
2022-02-28 04:57:50 +08:00
p_output < < INDENT2 " return " < < im_call < < " ( " < < icall_params < < " ); \n " ;
2017-10-29 09:37:13 +08:00
} else {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
p_output . append ( sformat ( return_type - > cs_out , im_call , icall_params ,
2022-02-28 04:57:50 +08:00
return_cs_type , return_type - > c_type_out , String ( ) , INDENT2 ) ) ;
2019-04-19 07:10:08 +08:00
p_output . append ( " \n " ) ;
2017-10-29 09:37:13 +08:00
}
2017-10-03 05:24:00 +08:00
2022-02-28 04:57:50 +08:00
p_output . append ( CLOSE_BLOCK_L1 ) ;
2017-10-29 09:37:13 +08:00
}
2017-10-03 05:24:00 +08:00
2017-10-29 09:37:13 +08:00
p_method_bind_count + + ;
2019-04-19 07:10:08 +08:00
2017-10-29 09:37:13 +08:00
return OK ;
}
2017-10-03 05:24:00 +08:00
2020-03-15 02:20:17 +08:00
Error BindingsGenerator : : _generate_cs_signal ( const BindingsGenerator : : TypeInterface & p_itype , const BindingsGenerator : : SignalInterface & p_isignal , StringBuilder & p_output ) {
String arguments_sig ;
C#: Remove need for reflection to invoking callable delegates
We aim to make the C# API reflection-free, mainly for concerns about
performance, and to be able to target NativeAOT in refletion-free mode,
which reduces the binary size.
One of the main usages of reflection still left was the dynamic
invokation of callable delegates, and for some time I wasn't sure
I would find an alternative solution that I'd be happy with.
The new solution uses trampoline functions to invoke the delegates:
```
static void Trampoline(object delegateObj, NativeVariantPtrArgs args, out godot_variant ret)
{
if (args.Count != 1)
throw new ArgumentException($"Callable expected 1 arguments but received {args.Count}.");
string res = ((Func<int, string>)delegateObj)(
VariantConversionCallbacks.GetToManagedCallback<int>()(args[0])
);
ret = VariantConversionCallbacks.GetToVariantCallback<string>()(res);
}
Callable.CreateWithUnsafeTrampoline((int num) => "Foo" + num, &Trampoline);
```
Of course, this is too much boilerplate for user code. To improve this,
the `Callable.From` methods were added. These are overloads that take
`Action` and `Func` delegates, which covers the most common use cases:
lambdas and method groups:
```
// Lambda
Callable.From((int num) => "Foo" + num);
// Method group
string AppendNum(int num) => "Foo" + num;
Callable.From(AppendNum);
```
Unfortunately, due to limitations in the C# language, implicit
conversions from delegates to `Callable` are not supported.
`Callable.From` does not support custom delegates. These should be
uncommon, but the Godot C# API actually uses them for event signals.
As such, the bindings generator was updated to generate trampoline
functions for event signals. It was also optimized to use `Action`
instead of a custom delegate for parameterless signals, which removes
the need for the trampoline functions for those signals.
The change to reflection-free invokation removes one of the last needs
for `ConvertVariantToManagedObjectOfType`. The only remaining usage is
from calling script constructors with parameters from the engine
(`CreateManagedForGodotObjectScriptInstance`). Once that one is made
reflection-free, `ConvertVariantToManagedObjectOfType` can be removed.
2022-10-29 04:59:13 +08:00
String delegate_type_params ;
if ( ! p_isignal . arguments . is_empty ( ) ) {
delegate_type_params + = " < " ;
}
2020-03-15 02:20:17 +08:00
// Retrieve information from the arguments
2021-07-26 00:45:00 +08:00
const ArgumentInterface & first = p_isignal . arguments . front ( ) - > get ( ) ;
for ( const ArgumentInterface & iarg : p_isignal . arguments ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
const TypeInterface * arg_type = _get_type_or_null ( iarg . type ) ;
ERR_FAIL_NULL_V ( arg_type , ERR_BUG ) ; // Argument type not found
2020-03-15 02:20:17 +08:00
2020-05-16 10:03:05 +08:00
ERR_FAIL_COND_V_MSG ( arg_type - > is_singleton , ERR_BUG ,
2021-01-09 07:14:25 +08:00
" Argument type is a singleton: ' " + iarg . name + " ' of signal ' " + p_itype . name + " . " + p_isignal . name + " '. " ) ;
if ( p_itype . api_type = = ClassDB : : API_CORE ) {
ERR_FAIL_COND_V_MSG ( arg_type - > api_type = = ClassDB : : API_EDITOR , ERR_BUG ,
" Argument ' " + iarg . name + " ' of signal ' " + p_itype . name + " . " + p_isignal . name + " ' has type ' " +
arg_type - > name + " ' from the editor API. Core API cannot have dependencies on the editor API. " ) ;
}
2020-05-16 10:03:05 +08:00
2020-03-15 02:20:17 +08:00
// Add the current arguments to the signature
2021-07-26 00:45:00 +08:00
if ( & iarg ! = & first ) {
2020-03-15 02:20:17 +08:00
arguments_sig + = " , " ;
C#: Remove need for reflection to invoking callable delegates
We aim to make the C# API reflection-free, mainly for concerns about
performance, and to be able to target NativeAOT in refletion-free mode,
which reduces the binary size.
One of the main usages of reflection still left was the dynamic
invokation of callable delegates, and for some time I wasn't sure
I would find an alternative solution that I'd be happy with.
The new solution uses trampoline functions to invoke the delegates:
```
static void Trampoline(object delegateObj, NativeVariantPtrArgs args, out godot_variant ret)
{
if (args.Count != 1)
throw new ArgumentException($"Callable expected 1 arguments but received {args.Count}.");
string res = ((Func<int, string>)delegateObj)(
VariantConversionCallbacks.GetToManagedCallback<int>()(args[0])
);
ret = VariantConversionCallbacks.GetToVariantCallback<string>()(res);
}
Callable.CreateWithUnsafeTrampoline((int num) => "Foo" + num, &Trampoline);
```
Of course, this is too much boilerplate for user code. To improve this,
the `Callable.From` methods were added. These are overloads that take
`Action` and `Func` delegates, which covers the most common use cases:
lambdas and method groups:
```
// Lambda
Callable.From((int num) => "Foo" + num);
// Method group
string AppendNum(int num) => "Foo" + num;
Callable.From(AppendNum);
```
Unfortunately, due to limitations in the C# language, implicit
conversions from delegates to `Callable` are not supported.
`Callable.From` does not support custom delegates. These should be
uncommon, but the Godot C# API actually uses them for event signals.
As such, the bindings generator was updated to generate trampoline
functions for event signals. It was also optimized to use `Action`
instead of a custom delegate for parameterless signals, which removes
the need for the trampoline functions for those signals.
The change to reflection-free invokation removes one of the last needs
for `ConvertVariantToManagedObjectOfType`. The only remaining usage is
from calling script constructors with parameters from the engine
(`CreateManagedForGodotObjectScriptInstance`). Once that one is made
reflection-free, `ConvertVariantToManagedObjectOfType` can be removed.
2022-10-29 04:59:13 +08:00
delegate_type_params + = " , " ;
2020-05-16 10:03:05 +08:00
}
2020-03-15 02:20:17 +08:00
arguments_sig + = arg_type - > cs_type ;
arguments_sig + = " " ;
arguments_sig + = iarg . name ;
C#: Remove need for reflection to invoking callable delegates
We aim to make the C# API reflection-free, mainly for concerns about
performance, and to be able to target NativeAOT in refletion-free mode,
which reduces the binary size.
One of the main usages of reflection still left was the dynamic
invokation of callable delegates, and for some time I wasn't sure
I would find an alternative solution that I'd be happy with.
The new solution uses trampoline functions to invoke the delegates:
```
static void Trampoline(object delegateObj, NativeVariantPtrArgs args, out godot_variant ret)
{
if (args.Count != 1)
throw new ArgumentException($"Callable expected 1 arguments but received {args.Count}.");
string res = ((Func<int, string>)delegateObj)(
VariantConversionCallbacks.GetToManagedCallback<int>()(args[0])
);
ret = VariantConversionCallbacks.GetToVariantCallback<string>()(res);
}
Callable.CreateWithUnsafeTrampoline((int num) => "Foo" + num, &Trampoline);
```
Of course, this is too much boilerplate for user code. To improve this,
the `Callable.From` methods were added. These are overloads that take
`Action` and `Func` delegates, which covers the most common use cases:
lambdas and method groups:
```
// Lambda
Callable.From((int num) => "Foo" + num);
// Method group
string AppendNum(int num) => "Foo" + num;
Callable.From(AppendNum);
```
Unfortunately, due to limitations in the C# language, implicit
conversions from delegates to `Callable` are not supported.
`Callable.From` does not support custom delegates. These should be
uncommon, but the Godot C# API actually uses them for event signals.
As such, the bindings generator was updated to generate trampoline
functions for event signals. It was also optimized to use `Action`
instead of a custom delegate for parameterless signals, which removes
the need for the trampoline functions for those signals.
The change to reflection-free invokation removes one of the last needs
for `ConvertVariantToManagedObjectOfType`. The only remaining usage is
from calling script constructors with parameters from the engine
(`CreateManagedForGodotObjectScriptInstance`). Once that one is made
reflection-free, `ConvertVariantToManagedObjectOfType` can be removed.
2022-10-29 04:59:13 +08:00
delegate_type_params + = arg_type - > cs_type ;
}
if ( ! p_isignal . arguments . is_empty ( ) ) {
delegate_type_params + = " > " ;
2020-03-15 02:20:17 +08:00
}
// Generate signal
{
2022-09-16 15:30:04 +08:00
p_output . append ( MEMBER_BEGIN " /// <summary> \n " ) ;
p_output . append ( INDENT1 " /// " ) ;
p_output . append ( " Represents the method that handles the " ) ;
p_output . append ( " <see cref= \" " BINDINGS_NAMESPACE " . " + p_itype . proxy_name + " . " + p_isignal . proxy_name + " \" /> " ) ;
p_output . append ( " event of a " ) ;
p_output . append ( " <see cref= \" " BINDINGS_NAMESPACE " . " + p_itype . proxy_name + " \" /> " ) ;
p_output . append ( " class. \n " ) ;
p_output . append ( INDENT1 " /// </summary> " ) ;
2022-09-15 19:24:41 +08:00
if ( p_isignal . is_deprecated ) {
if ( p_isignal . deprecation_message . is_empty ( ) ) {
WARN_PRINT ( " An empty deprecation message is discouraged. Signal: ' " + p_isignal . proxy_name + " '. " ) ;
}
p_output . append ( MEMBER_BEGIN " [Obsolete( \" " ) ;
p_output . append ( p_isignal . deprecation_message ) ;
p_output . append ( " \" )] " ) ;
}
C#: Remove need for reflection to invoking callable delegates
We aim to make the C# API reflection-free, mainly for concerns about
performance, and to be able to target NativeAOT in refletion-free mode,
which reduces the binary size.
One of the main usages of reflection still left was the dynamic
invokation of callable delegates, and for some time I wasn't sure
I would find an alternative solution that I'd be happy with.
The new solution uses trampoline functions to invoke the delegates:
```
static void Trampoline(object delegateObj, NativeVariantPtrArgs args, out godot_variant ret)
{
if (args.Count != 1)
throw new ArgumentException($"Callable expected 1 arguments but received {args.Count}.");
string res = ((Func<int, string>)delegateObj)(
VariantConversionCallbacks.GetToManagedCallback<int>()(args[0])
);
ret = VariantConversionCallbacks.GetToVariantCallback<string>()(res);
}
Callable.CreateWithUnsafeTrampoline((int num) => "Foo" + num, &Trampoline);
```
Of course, this is too much boilerplate for user code. To improve this,
the `Callable.From` methods were added. These are overloads that take
`Action` and `Func` delegates, which covers the most common use cases:
lambdas and method groups:
```
// Lambda
Callable.From((int num) => "Foo" + num);
// Method group
string AppendNum(int num) => "Foo" + num;
Callable.From(AppendNum);
```
Unfortunately, due to limitations in the C# language, implicit
conversions from delegates to `Callable` are not supported.
`Callable.From` does not support custom delegates. These should be
uncommon, but the Godot C# API actually uses them for event signals.
As such, the bindings generator was updated to generate trampoline
functions for event signals. It was also optimized to use `Action`
instead of a custom delegate for parameterless signals, which removes
the need for the trampoline functions for those signals.
The change to reflection-free invokation removes one of the last needs
for `ConvertVariantToManagedObjectOfType`. The only remaining usage is
from calling script constructors with parameters from the engine
(`CreateManagedForGodotObjectScriptInstance`). Once that one is made
reflection-free, `ConvertVariantToManagedObjectOfType` can be removed.
2022-10-29 04:59:13 +08:00
bool is_parameterless = p_isignal . arguments . size ( ) = = 0 ;
2022-09-15 19:24:41 +08:00
C#: Remove need for reflection to invoking callable delegates
We aim to make the C# API reflection-free, mainly for concerns about
performance, and to be able to target NativeAOT in refletion-free mode,
which reduces the binary size.
One of the main usages of reflection still left was the dynamic
invokation of callable delegates, and for some time I wasn't sure
I would find an alternative solution that I'd be happy with.
The new solution uses trampoline functions to invoke the delegates:
```
static void Trampoline(object delegateObj, NativeVariantPtrArgs args, out godot_variant ret)
{
if (args.Count != 1)
throw new ArgumentException($"Callable expected 1 arguments but received {args.Count}.");
string res = ((Func<int, string>)delegateObj)(
VariantConversionCallbacks.GetToManagedCallback<int>()(args[0])
);
ret = VariantConversionCallbacks.GetToVariantCallback<string>()(res);
}
Callable.CreateWithUnsafeTrampoline((int num) => "Foo" + num, &Trampoline);
```
Of course, this is too much boilerplate for user code. To improve this,
the `Callable.From` methods were added. These are overloads that take
`Action` and `Func` delegates, which covers the most common use cases:
lambdas and method groups:
```
// Lambda
Callable.From((int num) => "Foo" + num);
// Method group
string AppendNum(int num) => "Foo" + num;
Callable.From(AppendNum);
```
Unfortunately, due to limitations in the C# language, implicit
conversions from delegates to `Callable` are not supported.
`Callable.From` does not support custom delegates. These should be
uncommon, but the Godot C# API actually uses them for event signals.
As such, the bindings generator was updated to generate trampoline
functions for event signals. It was also optimized to use `Action`
instead of a custom delegate for parameterless signals, which removes
the need for the trampoline functions for those signals.
The change to reflection-free invokation removes one of the last needs
for `ConvertVariantToManagedObjectOfType`. The only remaining usage is
from calling script constructors with parameters from the engine
(`CreateManagedForGodotObjectScriptInstance`). Once that one is made
reflection-free, `ConvertVariantToManagedObjectOfType` can be removed.
2022-10-29 04:59:13 +08:00
// Delegate name is [SignalName]EventHandler
String delegate_name = is_parameterless ? " Action " : p_isignal . proxy_name + " EventHandler " ;
if ( ! is_parameterless ) {
// Generate delegate
p_output . append ( MEMBER_BEGIN " public delegate void " ) ;
p_output . append ( delegate_name ) ;
p_output . append ( " ( " ) ;
p_output . append ( arguments_sig ) ;
p_output . append ( " ); \n " ) ;
// Generate Callable trampoline for the delegate
C#: Optimize Variant conversion callbacks
These callbacks are used for marshaling by callables and generic Godot
collections.
C# generics don't support specialization the way C++ templates do.
I knew NativeAOT could optimize away many type checks when the types
are known at compile time, but I didn't trust the JIT would do as good
a job, so I initially went with cached function pointers.
Well, it turns out the JIT is also very good at optimizing in this
scenario, so I'm changing the methods to do the conversion directly,
rather than returning a function pointer for the conversion.
The methods were moved to `VariantUtils`, and were renamed from
`GetFromVariantCallback/GetToVariantCallback` to `ConvertTo/CreateFrom`.
The new implementation looks like it goes through many `if` checks
at runtime to find the right branch for the type, but in practice it
works pretty much like template specialization. The JIT only generates
code for the relevant branch. Together with inlining, the result is
very close or the same as doing the conversion manually:
```cs
godot_variant variant;
int foo = variant.Int;
int bar = VariantUtils.ConvertTo<int>(variant);
```
If the type is a generic Godot collection, the conversion still goes
through a function pointer call.
The new code happens to be much shorter as well, with the file going
from 1057 lines to 407.
Side note: `Variant.cs` was mistakenly created in the wrong folder,
so I moved it to the `Core` folder.
2022-11-06 08:27:55 +08:00
p_output < < MEMBER_BEGIN " private static void " < < p_isignal . proxy_name < < " Trampoline "
C#: Remove need for reflection to invoking callable delegates
We aim to make the C# API reflection-free, mainly for concerns about
performance, and to be able to target NativeAOT in refletion-free mode,
which reduces the binary size.
One of the main usages of reflection still left was the dynamic
invokation of callable delegates, and for some time I wasn't sure
I would find an alternative solution that I'd be happy with.
The new solution uses trampoline functions to invoke the delegates:
```
static void Trampoline(object delegateObj, NativeVariantPtrArgs args, out godot_variant ret)
{
if (args.Count != 1)
throw new ArgumentException($"Callable expected 1 arguments but received {args.Count}.");
string res = ((Func<int, string>)delegateObj)(
VariantConversionCallbacks.GetToManagedCallback<int>()(args[0])
);
ret = VariantConversionCallbacks.GetToVariantCallback<string>()(res);
}
Callable.CreateWithUnsafeTrampoline((int num) => "Foo" + num, &Trampoline);
```
Of course, this is too much boilerplate for user code. To improve this,
the `Callable.From` methods were added. These are overloads that take
`Action` and `Func` delegates, which covers the most common use cases:
lambdas and method groups:
```
// Lambda
Callable.From((int num) => "Foo" + num);
// Method group
string AppendNum(int num) => "Foo" + num;
Callable.From(AppendNum);
```
Unfortunately, due to limitations in the C# language, implicit
conversions from delegates to `Callable` are not supported.
`Callable.From` does not support custom delegates. These should be
uncommon, but the Godot C# API actually uses them for event signals.
As such, the bindings generator was updated to generate trampoline
functions for event signals. It was also optimized to use `Action`
instead of a custom delegate for parameterless signals, which removes
the need for the trampoline functions for those signals.
The change to reflection-free invokation removes one of the last needs
for `ConvertVariantToManagedObjectOfType`. The only remaining usage is
from calling script constructors with parameters from the engine
(`CreateManagedForGodotObjectScriptInstance`). Once that one is made
reflection-free, `ConvertVariantToManagedObjectOfType` can be removed.
2022-10-29 04:59:13 +08:00
< < " (object delegateObj, NativeVariantPtrArgs args, out godot_variant ret) \n "
< < INDENT1 " { \n "
< < INDENT2 " Callable.ThrowIfArgCountMismatch(args, " < < itos ( p_isignal . arguments . size ( ) ) < < " ); \n "
< < INDENT2 " (( " < < delegate_name < < " )delegateObj)( " ;
int idx = 0 ;
for ( const ArgumentInterface & iarg : p_isignal . arguments ) {
const TypeInterface * arg_type = _get_type_or_null ( iarg . type ) ;
ERR_FAIL_NULL_V ( arg_type , ERR_BUG ) ; // Argument type not found
if ( idx ! = 0 ) {
p_output < < " , " ;
}
C#: Optimize Variant conversion callbacks
These callbacks are used for marshaling by callables and generic Godot
collections.
C# generics don't support specialization the way C++ templates do.
I knew NativeAOT could optimize away many type checks when the types
are known at compile time, but I didn't trust the JIT would do as good
a job, so I initially went with cached function pointers.
Well, it turns out the JIT is also very good at optimizing in this
scenario, so I'm changing the methods to do the conversion directly,
rather than returning a function pointer for the conversion.
The methods were moved to `VariantUtils`, and were renamed from
`GetFromVariantCallback/GetToVariantCallback` to `ConvertTo/CreateFrom`.
The new implementation looks like it goes through many `if` checks
at runtime to find the right branch for the type, but in practice it
works pretty much like template specialization. The JIT only generates
code for the relevant branch. Together with inlining, the result is
very close or the same as doing the conversion manually:
```cs
godot_variant variant;
int foo = variant.Int;
int bar = VariantUtils.ConvertTo<int>(variant);
```
If the type is a generic Godot collection, the conversion still goes
through a function pointer call.
The new code happens to be much shorter as well, with the file going
from 1057 lines to 407.
Side note: `Variant.cs` was mistakenly created in the wrong folder,
so I moved it to the `Core` folder.
2022-11-06 08:27:55 +08:00
p_output < < sformat ( arg_type - > cs_variant_to_managed ,
" args[ " + itos ( idx ) + " ] " , arg_type - > cs_type , arg_type - > name ) ;
C#: Remove need for reflection to invoking callable delegates
We aim to make the C# API reflection-free, mainly for concerns about
performance, and to be able to target NativeAOT in refletion-free mode,
which reduces the binary size.
One of the main usages of reflection still left was the dynamic
invokation of callable delegates, and for some time I wasn't sure
I would find an alternative solution that I'd be happy with.
The new solution uses trampoline functions to invoke the delegates:
```
static void Trampoline(object delegateObj, NativeVariantPtrArgs args, out godot_variant ret)
{
if (args.Count != 1)
throw new ArgumentException($"Callable expected 1 arguments but received {args.Count}.");
string res = ((Func<int, string>)delegateObj)(
VariantConversionCallbacks.GetToManagedCallback<int>()(args[0])
);
ret = VariantConversionCallbacks.GetToVariantCallback<string>()(res);
}
Callable.CreateWithUnsafeTrampoline((int num) => "Foo" + num, &Trampoline);
```
Of course, this is too much boilerplate for user code. To improve this,
the `Callable.From` methods were added. These are overloads that take
`Action` and `Func` delegates, which covers the most common use cases:
lambdas and method groups:
```
// Lambda
Callable.From((int num) => "Foo" + num);
// Method group
string AppendNum(int num) => "Foo" + num;
Callable.From(AppendNum);
```
Unfortunately, due to limitations in the C# language, implicit
conversions from delegates to `Callable` are not supported.
`Callable.From` does not support custom delegates. These should be
uncommon, but the Godot C# API actually uses them for event signals.
As such, the bindings generator was updated to generate trampoline
functions for event signals. It was also optimized to use `Action`
instead of a custom delegate for parameterless signals, which removes
the need for the trampoline functions for those signals.
The change to reflection-free invokation removes one of the last needs
for `ConvertVariantToManagedObjectOfType`. The only remaining usage is
from calling script constructors with parameters from the engine
(`CreateManagedForGodotObjectScriptInstance`). Once that one is made
reflection-free, `ConvertVariantToManagedObjectOfType` can be removed.
2022-10-29 04:59:13 +08:00
idx + + ;
}
p_output < < " ); \n "
< < INDENT2 " ret = default; \n "
< < INDENT1 " } \n " ;
}
2022-09-15 19:24:41 +08:00
2020-03-15 02:20:17 +08:00
if ( p_isignal . method_doc & & p_isignal . method_doc - > description . size ( ) ) {
String xml_summary = bbcode_to_xml ( fix_doc_description ( p_isignal . method_doc - > description ) , & p_itype ) ;
Vector < String > summary_lines = xml_summary . length ( ) ? xml_summary . split ( " \n " ) : Vector < String > ( ) ;
if ( summary_lines . size ( ) ) {
p_output . append ( MEMBER_BEGIN " /// <summary> \n " ) ;
for ( int i = 0 ; i < summary_lines . size ( ) ; i + + ) {
2022-02-28 04:57:50 +08:00
p_output . append ( INDENT1 " /// " ) ;
2020-03-15 02:20:17 +08:00
p_output . append ( summary_lines [ i ] ) ;
p_output . append ( " \n " ) ;
}
2022-02-28 04:57:50 +08:00
p_output . append ( INDENT1 " /// </summary> " ) ;
2020-03-15 02:20:17 +08:00
}
}
if ( p_isignal . is_deprecated ) {
p_output . append ( MEMBER_BEGIN " [Obsolete( \" " ) ;
p_output . append ( p_isignal . deprecation_message ) ;
p_output . append ( " \" )] " ) ;
}
// TODO:
// Could we assume the StringName instance of signal name will never be freed (it's stored in ClassDB) before the managed world is unloaded?
// If so, we could store the pointer we get from `data_unique_pointer()` instead of allocating StringName here.
// Generate event
2022-08-26 06:49:14 +08:00
p_output . append ( MEMBER_BEGIN " public " ) ;
2020-03-15 02:20:17 +08:00
2020-05-16 10:03:05 +08:00
if ( p_itype . is_singleton ) {
2020-03-15 02:20:17 +08:00
p_output . append ( " static " ) ;
2020-05-16 10:03:05 +08:00
}
2020-03-15 02:20:17 +08:00
C#: Remove need for reflection to invoking callable delegates
We aim to make the C# API reflection-free, mainly for concerns about
performance, and to be able to target NativeAOT in refletion-free mode,
which reduces the binary size.
One of the main usages of reflection still left was the dynamic
invokation of callable delegates, and for some time I wasn't sure
I would find an alternative solution that I'd be happy with.
The new solution uses trampoline functions to invoke the delegates:
```
static void Trampoline(object delegateObj, NativeVariantPtrArgs args, out godot_variant ret)
{
if (args.Count != 1)
throw new ArgumentException($"Callable expected 1 arguments but received {args.Count}.");
string res = ((Func<int, string>)delegateObj)(
VariantConversionCallbacks.GetToManagedCallback<int>()(args[0])
);
ret = VariantConversionCallbacks.GetToVariantCallback<string>()(res);
}
Callable.CreateWithUnsafeTrampoline((int num) => "Foo" + num, &Trampoline);
```
Of course, this is too much boilerplate for user code. To improve this,
the `Callable.From` methods were added. These are overloads that take
`Action` and `Func` delegates, which covers the most common use cases:
lambdas and method groups:
```
// Lambda
Callable.From((int num) => "Foo" + num);
// Method group
string AppendNum(int num) => "Foo" + num;
Callable.From(AppendNum);
```
Unfortunately, due to limitations in the C# language, implicit
conversions from delegates to `Callable` are not supported.
`Callable.From` does not support custom delegates. These should be
uncommon, but the Godot C# API actually uses them for event signals.
As such, the bindings generator was updated to generate trampoline
functions for event signals. It was also optimized to use `Action`
instead of a custom delegate for parameterless signals, which removes
the need for the trampoline functions for those signals.
The change to reflection-free invokation removes one of the last needs
for `ConvertVariantToManagedObjectOfType`. The only remaining usage is
from calling script constructors with parameters from the engine
(`CreateManagedForGodotObjectScriptInstance`). Once that one is made
reflection-free, `ConvertVariantToManagedObjectOfType` can be removed.
2022-10-29 04:59:13 +08:00
if ( ! is_parameterless ) {
// `unsafe` is needed for taking the trampoline's function pointer
p_output < < " unsafe " ;
}
2020-03-15 02:20:17 +08:00
p_output . append ( " event " ) ;
p_output . append ( delegate_name ) ;
p_output . append ( " " ) ;
p_output . append ( p_isignal . proxy_name ) ;
2022-02-28 04:57:50 +08:00
p_output . append ( " \n " OPEN_BLOCK_L1 INDENT2 ) ;
2020-03-15 02:20:17 +08:00
2020-05-16 10:03:05 +08:00
if ( p_itype . is_singleton ) {
2022-08-25 23:07:20 +08:00
p_output . append ( " add => " CS_PROPERTY_SINGLETON " .Connect(SignalName. " ) ;
2020-05-16 10:03:05 +08:00
} else {
2022-08-25 23:07:20 +08:00
p_output . append ( " add => Connect(SignalName. " ) ;
2020-05-16 10:03:05 +08:00
}
2020-03-15 02:20:17 +08:00
C#: Remove need for reflection to invoking callable delegates
We aim to make the C# API reflection-free, mainly for concerns about
performance, and to be able to target NativeAOT in refletion-free mode,
which reduces the binary size.
One of the main usages of reflection still left was the dynamic
invokation of callable delegates, and for some time I wasn't sure
I would find an alternative solution that I'd be happy with.
The new solution uses trampoline functions to invoke the delegates:
```
static void Trampoline(object delegateObj, NativeVariantPtrArgs args, out godot_variant ret)
{
if (args.Count != 1)
throw new ArgumentException($"Callable expected 1 arguments but received {args.Count}.");
string res = ((Func<int, string>)delegateObj)(
VariantConversionCallbacks.GetToManagedCallback<int>()(args[0])
);
ret = VariantConversionCallbacks.GetToVariantCallback<string>()(res);
}
Callable.CreateWithUnsafeTrampoline((int num) => "Foo" + num, &Trampoline);
```
Of course, this is too much boilerplate for user code. To improve this,
the `Callable.From` methods were added. These are overloads that take
`Action` and `Func` delegates, which covers the most common use cases:
lambdas and method groups:
```
// Lambda
Callable.From((int num) => "Foo" + num);
// Method group
string AppendNum(int num) => "Foo" + num;
Callable.From(AppendNum);
```
Unfortunately, due to limitations in the C# language, implicit
conversions from delegates to `Callable` are not supported.
`Callable.From` does not support custom delegates. These should be
uncommon, but the Godot C# API actually uses them for event signals.
As such, the bindings generator was updated to generate trampoline
functions for event signals. It was also optimized to use `Action`
instead of a custom delegate for parameterless signals, which removes
the need for the trampoline functions for those signals.
The change to reflection-free invokation removes one of the last needs
for `ConvertVariantToManagedObjectOfType`. The only remaining usage is
from calling script constructors with parameters from the engine
(`CreateManagedForGodotObjectScriptInstance`). Once that one is made
reflection-free, `ConvertVariantToManagedObjectOfType` can be removed.
2022-10-29 04:59:13 +08:00
if ( is_parameterless ) {
// Delegate type is Action. No need for custom trampoline.
p_output < < p_isignal . proxy_name < < " , Callable.From(value)); \n " ;
} else {
p_output < < p_isignal . proxy_name
< < " , Callable.CreateWithUnsafeTrampoline(value, & " < < p_isignal . proxy_name < < " Trampoline)); \n " ;
}
2020-03-15 02:20:17 +08:00
2020-05-16 10:03:05 +08:00
if ( p_itype . is_singleton ) {
2022-08-25 23:07:20 +08:00
p_output . append ( INDENT2 " remove => " CS_PROPERTY_SINGLETON " .Disconnect(SignalName. " ) ;
2020-05-16 10:03:05 +08:00
} else {
2022-08-25 23:07:20 +08:00
p_output . append ( INDENT2 " remove => Disconnect(SignalName. " ) ;
2020-05-16 10:03:05 +08:00
}
2020-03-15 02:20:17 +08:00
C#: Remove need for reflection to invoking callable delegates
We aim to make the C# API reflection-free, mainly for concerns about
performance, and to be able to target NativeAOT in refletion-free mode,
which reduces the binary size.
One of the main usages of reflection still left was the dynamic
invokation of callable delegates, and for some time I wasn't sure
I would find an alternative solution that I'd be happy with.
The new solution uses trampoline functions to invoke the delegates:
```
static void Trampoline(object delegateObj, NativeVariantPtrArgs args, out godot_variant ret)
{
if (args.Count != 1)
throw new ArgumentException($"Callable expected 1 arguments but received {args.Count}.");
string res = ((Func<int, string>)delegateObj)(
VariantConversionCallbacks.GetToManagedCallback<int>()(args[0])
);
ret = VariantConversionCallbacks.GetToVariantCallback<string>()(res);
}
Callable.CreateWithUnsafeTrampoline((int num) => "Foo" + num, &Trampoline);
```
Of course, this is too much boilerplate for user code. To improve this,
the `Callable.From` methods were added. These are overloads that take
`Action` and `Func` delegates, which covers the most common use cases:
lambdas and method groups:
```
// Lambda
Callable.From((int num) => "Foo" + num);
// Method group
string AppendNum(int num) => "Foo" + num;
Callable.From(AppendNum);
```
Unfortunately, due to limitations in the C# language, implicit
conversions from delegates to `Callable` are not supported.
`Callable.From` does not support custom delegates. These should be
uncommon, but the Godot C# API actually uses them for event signals.
As such, the bindings generator was updated to generate trampoline
functions for event signals. It was also optimized to use `Action`
instead of a custom delegate for parameterless signals, which removes
the need for the trampoline functions for those signals.
The change to reflection-free invokation removes one of the last needs
for `ConvertVariantToManagedObjectOfType`. The only remaining usage is
from calling script constructors with parameters from the engine
(`CreateManagedForGodotObjectScriptInstance`). Once that one is made
reflection-free, `ConvertVariantToManagedObjectOfType` can be removed.
2022-10-29 04:59:13 +08:00
if ( is_parameterless ) {
// Delegate type is Action. No need for custom trampoline.
p_output < < p_isignal . proxy_name < < " , Callable.From(value)); \n " ;
} else {
p_output < < p_isignal . proxy_name
< < " , Callable.CreateWithUnsafeTrampoline(value, & " < < p_isignal . proxy_name < < " Trampoline)); \n " ;
}
2022-02-28 04:57:50 +08:00
p_output . append ( CLOSE_BLOCK_L1 ) ;
2020-03-15 02:20:17 +08:00
}
return OK ;
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
Error BindingsGenerator : : _generate_cs_native_calls ( const InternalCall & p_icall , StringBuilder & r_output ) {
bool ret_void = p_icall . return_type . cname = = name_cache . type_void ;
2018-09-04 11:40:41 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
const TypeInterface * return_type = _get_type_or_null ( p_icall . return_type ) ;
ERR_FAIL_NULL_V ( return_type , ERR_BUG ) ; // Return type not found
2017-10-03 05:24:00 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
StringBuilder c_func_sig ;
StringBuilder c_in_statements ;
StringBuilder c_args_var_content ;
2017-10-03 05:24:00 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
c_func_sig < < " IntPtr " CS_PARAM_METHODBIND ;
2017-10-03 05:24:00 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
if ( ! p_icall . is_static ) {
c_func_sig + = " , IntPtr " CS_PARAM_INSTANCE ;
2017-10-03 05:24:00 +08:00
}
2017-10-29 09:37:13 +08:00
// Get arguments information
int i = 0 ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
for ( const TypeReference & arg_type_ref : p_icall . argument_types ) {
const TypeInterface * arg_type = _get_type_or_null ( arg_type_ref ) ;
ERR_FAIL_NULL_V ( arg_type , ERR_BUG ) ; // Return type not found
2017-10-29 09:37:13 +08:00
String c_param_name = " arg " + itos ( i + 1 ) ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
if ( p_icall . is_vararg ) {
if ( i < p_icall . get_arguments_count ( ) - 1 ) {
2021-12-29 06:25:16 +08:00
String c_in_vararg = arg_type - > c_in_vararg ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
if ( arg_type - > is_object_type ) {
2022-02-28 04:57:46 +08:00
c_in_vararg = " %5using godot_variant %1_in = VariantUtils.CreateFromGodotObjectPtr(%1); \n " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
}
ERR_FAIL_COND_V_MSG ( c_in_vararg . is_empty ( ) , ERR_BUG ,
" VarArg support not implemented for parameter type: " + arg_type - > name ) ;
c_in_statements
< < sformat ( c_in_vararg , return_type - > c_type , c_param_name ,
2022-02-28 04:57:50 +08:00
String ( ) , String ( ) , String ( ) , INDENT3 )
< < INDENT3 C_LOCAL_PTRCALL_ARGS " [ " < < itos ( i )
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
< < " ] = new IntPtr(& " < < c_param_name < < " _in); \n " ;
2017-10-29 09:37:13 +08:00
}
} else {
2020-05-16 10:03:05 +08:00
if ( i > 0 ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
c_args_var_content < < " , " ;
2020-05-16 10:03:05 +08:00
}
if ( arg_type - > c_in . size ( ) ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
c_in_statements < < sformat ( arg_type - > c_in , arg_type - > c_type , c_param_name ,
2022-02-28 04:57:50 +08:00
String ( ) , String ( ) , String ( ) , INDENT2 ) ;
2020-05-16 10:03:05 +08:00
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
c_args_var_content < < sformat ( arg_type - > c_arg_in , c_param_name ) ;
2017-10-29 09:37:13 +08:00
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
c_func_sig < < " , " < < arg_type - > c_type_in < < " " < < c_param_name ;
2017-10-29 09:37:13 +08:00
i + + ;
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
String icall_method = p_icall . name ;
2019-11-15 04:48:49 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
// Generate icall function
2017-10-29 09:37:13 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
r_output < < MEMBER_BEGIN " internal static unsafe " < < ( ret_void ? " void " : return_type - > c_type_out ) < < " "
< < icall_method < < " ( " < < c_func_sig . as_string ( ) < < " ) " OPEN_BLOCK ;
2017-10-29 09:37:13 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
if ( ! ret_void & & ( ! p_icall . is_vararg | | return_type - > cname ! = name_cache . type_Variant ) ) {
String ptrcall_return_type ;
String initialization ;
2017-10-29 09:37:13 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
if ( return_type - > is_object_type ) {
ptrcall_return_type = return_type - > is_ref_counted ? " godot_ref " : return_type - > c_type ;
initialization = " = default " ;
} else {
ptrcall_return_type = return_type - > c_type ;
2020-05-16 10:03:05 +08:00
}
2017-10-29 09:37:13 +08:00
2022-02-28 04:57:50 +08:00
r_output < < INDENT2 ;
2017-10-29 09:37:13 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
if ( return_type - > is_ref_counted | | return_type - > c_type_is_disposable_struct ) {
r_output < < " using " ;
2017-10-29 09:37:13 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
if ( initialization . is_empty ( ) ) {
initialization = " = default " ;
2019-01-18 07:41:41 +08:00
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
} else if ( return_type - > c_ret_needs_default_initialization ) {
initialization = " = default " ;
}
2019-01-18 07:41:41 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
r_output < < ptrcall_return_type < < " " C_LOCAL_RET < < initialization < < " ; \n " ;
}
2017-10-29 09:37:13 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
if ( ! p_icall . is_static ) {
2022-02-28 04:57:50 +08:00
r_output < < INDENT2 " if ( " CS_PARAM_INSTANCE " == IntPtr.Zero) \n "
< < INDENT3 " throw new ArgumentNullException(nameof( " CS_PARAM_INSTANCE " )); \n " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
}
2019-11-15 04:48:49 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
String argc_str = itos ( p_icall . get_arguments_count ( ) ) ;
2019-11-15 04:48:49 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
auto generate_call_and_return_stmts = [ & ] ( const char * base_indent ) {
if ( p_icall . is_vararg ) {
// MethodBind Call
r_output < < base_indent ;
2017-10-29 09:37:13 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
// VarArg methods always return Variant, but there are some cases in which MethodInfo provides
// a specific return type. We trust this information is valid. We need a temporary local to keep
// the Variant alive until the method returns. Otherwise, if the returned Variant holds a RefPtr,
// it could be deleted too early. This is the case with GDScript.new() which returns OBJECT.
// Alternatively, we could just return Variant, but that would result in a worse API.
2017-10-29 09:37:13 +08:00
2019-01-18 07:41:41 +08:00
if ( ! ret_void ) {
if ( return_type - > cname ! = name_cache . type_Variant ) {
2022-12-13 00:54:21 +08:00
// Usually the return value takes ownership, but in this case the variant is only used
// for conversion to another return type. As such, the local variable takes ownership.
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
r_output < < " using godot_variant " < < C_LOCAL_VARARG_RET " = " ;
2019-01-18 07:41:41 +08:00
} else {
2022-12-13 00:54:21 +08:00
// Variant's [c_out] takes ownership of the variant value
r_output < < " godot_variant " < < C_LOCAL_RET " = " ;
2019-01-18 07:41:41 +08:00
}
}
2017-10-29 09:37:13 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
r_output < < C_CLASS_NATIVE_FUNCS " .godotsharp_method_bind_call( "
< < CS_PARAM_METHODBIND " , " < < ( p_icall . is_static ? " IntPtr.Zero " : CS_PARAM_INSTANCE )
< < " , " < < ( p_icall . get_arguments_count ( ) ? " (godot_variant**) " C_LOCAL_PTRCALL_ARGS : " null " )
2021-12-29 06:25:16 +08:00
< < " , total_length, out _); \n " ;
2019-01-18 07:41:41 +08:00
2020-01-03 01:02:45 +08:00
if ( ! ret_void ) {
if ( return_type - > cname ! = name_cache . type_Variant ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
if ( return_type - > cname = = name_cache . enum_Error ) {
2021-12-29 06:25:16 +08:00
r_output < < base_indent < < C_LOCAL_RET " = VariantUtils.ConvertToInt64( " C_LOCAL_VARARG_RET " ); \n " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
} else {
// TODO: Use something similar to c_in_vararg (see usage above, with error if not implemented)
CRASH_NOW_MSG ( " Custom VarArg return type not implemented: " + return_type - > name ) ;
r_output < < base_indent < < C_LOCAL_RET " = " C_LOCAL_VARARG_RET " ; \n " ;
}
2020-01-03 01:02:45 +08:00
}
2019-01-18 07:41:41 +08:00
}
2017-10-29 09:37:13 +08:00
} else {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
// MethodBind PtrCall
r_output < < base_indent < < C_CLASS_NATIVE_FUNCS " .godotsharp_method_bind_ptrcall( "
< < CS_PARAM_METHODBIND " , " < < ( p_icall . is_static ? " IntPtr.Zero " : CS_PARAM_INSTANCE )
< < " , " < < ( p_icall . get_arguments_count ( ) ? C_LOCAL_PTRCALL_ARGS : " null " )
< < " , " < < ( ! ret_void ? " & " C_LOCAL_RET " ); \n " : " null); \n " ) ;
2017-10-29 09:37:13 +08:00
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
// Return statement
2017-10-29 09:37:13 +08:00
if ( ! ret_void ) {
2020-12-15 20:04:21 +08:00
if ( return_type - > c_out . is_empty ( ) ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
r_output < < base_indent < < " return " C_LOCAL_RET " ; \n " ;
2019-11-15 04:48:49 +08:00
} else {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
r_output < < sformat ( return_type - > c_out , return_type - > c_type_out , C_LOCAL_RET ,
return_type - > name , String ( ) , String ( ) , base_indent ) ;
2019-11-15 04:48:49 +08:00
}
2017-10-29 09:37:13 +08:00
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
} ;
if ( p_icall . get_arguments_count ( ) ) {
if ( p_icall . is_vararg ) {
String vararg_arg = " arg " + argc_str ;
String real_argc_str = itos ( p_icall . get_arguments_count ( ) - 1 ) ; // Arguments count without vararg
p_icall . get_arguments_count ( ) ;
2022-02-28 04:57:50 +08:00
r_output < < INDENT2 " int vararg_length = " < < vararg_arg < < " .Length; \n "
< < INDENT2 " int total_length = " < < real_argc_str < < " + vararg_length; \n " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
2022-02-28 04:57:50 +08:00
r_output < < INDENT2 " Span<godot_variant.movable> varargs_span = vararg_length <= VarArgsSpanThreshold ? \n "
2022-11-26 03:27:36 +08:00
< < INDENT3 " stackalloc godot_variant.movable[VarArgsSpanThreshold] : \n "
2022-02-28 04:57:50 +08:00
< < INDENT3 " new godot_variant.movable[vararg_length]; \n " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
2022-02-28 04:57:50 +08:00
r_output < < INDENT2 " Span<IntPtr> " C_LOCAL_PTRCALL_ARGS " _span = total_length <= VarArgsSpanThreshold ? \n "
< < INDENT3 " stackalloc IntPtr[VarArgsSpanThreshold] : \n "
< < INDENT3 " new IntPtr[total_length]; \n " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
2022-07-28 23:41:49 +08:00
r_output < < INDENT2 " fixed (godot_variant.movable* varargs = &MemoryMarshal.GetReference(varargs_span)) \n "
2022-02-28 04:57:50 +08:00
< < INDENT2 " fixed (IntPtr* " C_LOCAL_PTRCALL_ARGS " = "
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
" &MemoryMarshal.GetReference( " C_LOCAL_PTRCALL_ARGS " _span)) \n "
2022-02-28 04:57:50 +08:00
< < OPEN_BLOCK_L2 ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
r_output < < c_in_statements . as_string ( ) ;
2017-10-29 09:37:13 +08:00
2022-02-28 04:57:50 +08:00
r_output < < INDENT3 " for (int i = 0; i < vararg_length; i++) " OPEN_BLOCK
2022-07-28 23:41:49 +08:00
< < INDENT4 " varargs[i] = " < < vararg_arg < < " [i].NativeVar; \n "
2022-02-28 04:57:50 +08:00
< < INDENT4 C_LOCAL_PTRCALL_ARGS " [ " < < real_argc_str < < " + i] = new IntPtr(&varargs[i]); \n "
< < CLOSE_BLOCK_L3 ;
2017-10-29 09:37:13 +08:00
2022-02-28 04:57:50 +08:00
generate_call_and_return_stmts ( INDENT3 ) ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
2022-02-28 04:57:50 +08:00
r_output < < CLOSE_BLOCK_L2 ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
} else {
r_output < < c_in_statements . as_string ( ) ;
2022-02-28 04:57:50 +08:00
r_output < < INDENT2 " void** " C_LOCAL_PTRCALL_ARGS " = stackalloc void*[ "
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
< < argc_str < < " ] { " < < c_args_var_content . as_string ( ) < < " }; \n " ;
2022-02-28 04:57:50 +08:00
generate_call_and_return_stmts ( INDENT2 ) ;
2020-05-16 10:03:05 +08:00
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
} else {
2022-02-28 04:57:50 +08:00
generate_call_and_return_stmts ( INDENT2 ) ;
2017-10-29 09:37:13 +08:00
}
2022-02-28 04:57:50 +08:00
r_output < < CLOSE_BLOCK_L1 ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
return OK ;
}
Error BindingsGenerator : : _save_file ( const String & p_path , const StringBuilder & p_content ) {
Ref < FileAccess > file = FileAccess : : open ( p_path , FileAccess : : WRITE ) ;
ERR_FAIL_COND_V_MSG ( file . is_null ( ) , ERR_FILE_CANT_WRITE , " Cannot open file: ' " + p_path + " '. " ) ;
file - > store_string ( p_content . as_string ( ) ) ;
2017-10-29 09:37:13 +08:00
return OK ;
}
2018-04-29 04:25:25 +08:00
const BindingsGenerator : : TypeInterface * BindingsGenerator : : _get_type_or_null ( const TypeReference & p_typeref ) {
2022-05-13 21:04:37 +08:00
HashMap < StringName , TypeInterface > : : ConstIterator builtin_type_match = builtin_types . find ( p_typeref . cname ) ;
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
if ( builtin_type_match ) {
2022-05-13 21:04:37 +08:00
return & builtin_type_match - > value ;
2020-05-16 10:03:05 +08:00
}
2017-12-24 10:17:48 +08:00
2022-05-13 21:04:37 +08:00
HashMap < StringName , TypeInterface > : : ConstIterator obj_type_match = obj_types . find ( p_typeref . cname ) ;
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
if ( obj_type_match ) {
2022-05-08 16:09:19 +08:00
return & obj_type_match - > value ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
2018-04-29 04:25:25 +08:00
if ( p_typeref . is_enum ) {
2022-05-13 21:04:37 +08:00
HashMap < StringName , TypeInterface > : : ConstIterator enum_match = enum_types . find ( p_typeref . cname ) ;
2017-10-03 05:24:00 +08:00
2020-05-16 10:03:05 +08:00
if ( enum_match ) {
2022-05-13 21:04:37 +08:00
return & enum_match - > value ;
2020-05-16 10:03:05 +08:00
}
2018-04-29 04:25:25 +08:00
// Enum not found. Most likely because none of its constants were bound, so it's empty. That's fine. Use int instead.
2022-05-13 21:04:37 +08:00
HashMap < StringName , TypeInterface > : : ConstIterator int_match = builtin_types . find ( name_cache . type_int ) ;
2020-04-02 07:20:12 +08:00
ERR_FAIL_NULL_V ( int_match , nullptr ) ;
2022-05-13 21:04:37 +08:00
return & int_match - > value ;
2018-04-29 04:25:25 +08:00
}
2017-10-03 05:24:00 +08:00
2020-04-02 07:20:12 +08:00
return nullptr ;
2017-10-03 05:24:00 +08:00
}
2022-02-09 07:56:10 +08:00
const String BindingsGenerator : : _get_generic_type_parameters ( const TypeInterface & p_itype , const List < TypeReference > & p_generic_type_parameters ) {
if ( p_generic_type_parameters . is_empty ( ) ) {
return " " ;
}
ERR_FAIL_COND_V_MSG ( p_itype . type_parameter_count ! = p_generic_type_parameters . size ( ) , " " ,
" Generic type parameter count mismatch for type ' " + p_itype . name + " '. " +
" Found " + itos ( p_generic_type_parameters . size ( ) ) + " , but requires " +
itos ( p_itype . type_parameter_count ) + " . " ) ;
int i = 0 ;
String params = " < " ;
for ( const TypeReference & param_type : p_generic_type_parameters ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
const TypeInterface * param_itype = _get_type_or_null ( param_type ) ;
ERR_FAIL_NULL_V ( param_itype , " " ) ;
2022-02-09 07:56:10 +08:00
ERR_FAIL_COND_V_MSG ( param_itype - > is_singleton , " " ,
" Generic type parameter is a singleton: ' " + param_itype - > name + " '. " ) ;
if ( p_itype . api_type = = ClassDB : : API_CORE ) {
ERR_FAIL_COND_V_MSG ( param_itype - > api_type = = ClassDB : : API_EDITOR , " " ,
" Generic type parameter ' " + param_itype - > name + " ' has type from the editor API. " +
" Core API cannot have dependencies on the editor API. " ) ;
}
params + = param_itype - > cs_type ;
if ( i < p_generic_type_parameters . size ( ) - 1 ) {
params + = " , " ;
}
i + + ;
}
params + = " > " ;
return params ;
}
2022-09-01 06:40:59 +08:00
StringName BindingsGenerator : : _get_type_name_from_meta ( Variant : : Type p_type , GodotTypeInfo : : Metadata p_meta ) {
if ( p_type = = Variant : : INT ) {
return _get_int_type_name_from_meta ( p_meta ) ;
} else if ( p_type = = Variant : : FLOAT ) {
return _get_float_type_name_from_meta ( p_meta ) ;
} else {
return Variant : : get_type_name ( p_type ) ;
}
}
2019-04-26 02:24:48 +08:00
StringName BindingsGenerator : : _get_int_type_name_from_meta ( GodotTypeInfo : : Metadata p_meta ) {
switch ( p_meta ) {
case GodotTypeInfo : : METADATA_INT_IS_INT8 :
return " sbyte " ;
break ;
case GodotTypeInfo : : METADATA_INT_IS_INT16 :
return " short " ;
break ;
case GodotTypeInfo : : METADATA_INT_IS_INT32 :
return " int " ;
break ;
case GodotTypeInfo : : METADATA_INT_IS_INT64 :
return " long " ;
break ;
case GodotTypeInfo : : METADATA_INT_IS_UINT8 :
return " byte " ;
break ;
case GodotTypeInfo : : METADATA_INT_IS_UINT16 :
return " ushort " ;
break ;
case GodotTypeInfo : : METADATA_INT_IS_UINT32 :
return " uint " ;
break ;
case GodotTypeInfo : : METADATA_INT_IS_UINT64 :
return " ulong " ;
break ;
default :
2022-09-01 06:40:59 +08:00
// Assume INT64
return " long " ;
2019-04-26 02:24:48 +08:00
}
}
StringName BindingsGenerator : : _get_float_type_name_from_meta ( GodotTypeInfo : : Metadata p_meta ) {
switch ( p_meta ) {
case GodotTypeInfo : : METADATA_REAL_IS_FLOAT :
return " float " ;
break ;
case GodotTypeInfo : : METADATA_REAL_IS_DOUBLE :
return " double " ;
break ;
default :
2022-09-01 06:40:59 +08:00
// Assume FLOAT64
2019-04-26 02:24:48 +08:00
return " double " ;
}
}
2020-05-16 10:03:05 +08:00
bool BindingsGenerator : : _arg_default_value_is_assignable_to_type ( const Variant & p_val , const TypeInterface & p_arg_type ) {
if ( p_arg_type . name = = name_cache . type_Variant ) {
// Variant can take anything
return true ;
}
switch ( p_val . get_type ( ) ) {
case Variant : : NIL :
return p_arg_type . is_object_type | |
2021-10-28 21:19:35 +08:00
name_cache . is_nullable_type ( p_arg_type . name ) ;
2020-05-16 10:03:05 +08:00
case Variant : : BOOL :
return p_arg_type . name = = name_cache . type_bool ;
case Variant : : INT :
return p_arg_type . name = = name_cache . type_sbyte | |
2021-10-28 21:19:35 +08:00
p_arg_type . name = = name_cache . type_short | |
p_arg_type . name = = name_cache . type_int | |
p_arg_type . name = = name_cache . type_byte | |
p_arg_type . name = = name_cache . type_ushort | |
p_arg_type . name = = name_cache . type_uint | |
p_arg_type . name = = name_cache . type_long | |
p_arg_type . name = = name_cache . type_ulong | |
p_arg_type . name = = name_cache . type_float | |
p_arg_type . name = = name_cache . type_double | |
p_arg_type . is_enum ;
2020-05-16 10:03:05 +08:00
case Variant : : FLOAT :
return p_arg_type . name = = name_cache . type_float | |
2021-10-28 21:19:35 +08:00
p_arg_type . name = = name_cache . type_double ;
2020-05-16 10:03:05 +08:00
case Variant : : STRING :
case Variant : : STRING_NAME :
return p_arg_type . name = = name_cache . type_String | |
2021-10-28 21:19:35 +08:00
p_arg_type . name = = name_cache . type_StringName | |
p_arg_type . name = = name_cache . type_NodePath ;
2020-05-16 10:03:05 +08:00
case Variant : : NODE_PATH :
return p_arg_type . name = = name_cache . type_NodePath ;
case Variant : : TRANSFORM2D :
2021-04-28 15:36:08 +08:00
case Variant : : TRANSFORM3D :
2020-05-16 10:03:05 +08:00
case Variant : : BASIS :
2021-01-20 15:02:02 +08:00
case Variant : : QUATERNION :
2020-05-16 10:03:05 +08:00
case Variant : : PLANE :
case Variant : : AABB :
case Variant : : COLOR :
case Variant : : VECTOR2 :
case Variant : : RECT2 :
case Variant : : VECTOR3 :
2020-11-09 21:53:05 +08:00
case Variant : : RID :
2020-05-16 10:03:05 +08:00
case Variant : : PACKED_BYTE_ARRAY :
case Variant : : PACKED_INT32_ARRAY :
case Variant : : PACKED_INT64_ARRAY :
case Variant : : PACKED_FLOAT32_ARRAY :
case Variant : : PACKED_FLOAT64_ARRAY :
case Variant : : PACKED_STRING_ARRAY :
case Variant : : PACKED_VECTOR2_ARRAY :
case Variant : : PACKED_VECTOR3_ARRAY :
case Variant : : PACKED_COLOR_ARRAY :
case Variant : : CALLABLE :
case Variant : : SIGNAL :
return p_arg_type . name = = Variant : : get_type_name ( p_val . get_type ( ) ) ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
case Variant : : ARRAY :
return p_arg_type . name = = Variant : : get_type_name ( p_val . get_type ( ) ) | | p_arg_type . cname = = name_cache . type_Array_generic ;
case Variant : : DICTIONARY :
return p_arg_type . name = = Variant : : get_type_name ( p_val . get_type ( ) ) | | p_arg_type . cname = = name_cache . type_Dictionary_generic ;
2020-05-16 10:03:05 +08:00
case Variant : : OBJECT :
return p_arg_type . is_object_type ;
case Variant : : VECTOR2I :
return p_arg_type . name = = name_cache . type_Vector2 | |
2021-10-28 21:19:35 +08:00
p_arg_type . name = = Variant : : get_type_name ( p_val . get_type ( ) ) ;
2020-05-16 10:03:05 +08:00
case Variant : : RECT2I :
return p_arg_type . name = = name_cache . type_Rect2 | |
2021-10-28 21:19:35 +08:00
p_arg_type . name = = Variant : : get_type_name ( p_val . get_type ( ) ) ;
2020-05-16 10:03:05 +08:00
case Variant : : VECTOR3I :
return p_arg_type . name = = name_cache . type_Vector3 | |
2021-10-28 21:19:35 +08:00
p_arg_type . name = = Variant : : get_type_name ( p_val . get_type ( ) ) ;
2020-05-16 10:03:05 +08:00
default :
CRASH_NOW_MSG ( " Unexpected Variant type: " + itos ( p_val . get_type ( ) ) ) ;
break ;
}
return false ;
}
2023-01-17 11:32:05 +08:00
bool method_has_ptr_parameter ( MethodInfo p_method_info ) {
if ( p_method_info . return_val . type = = Variant : : INT & & p_method_info . return_val . hint = = PROPERTY_HINT_INT_IS_POINTER ) {
return true ;
}
for ( PropertyInfo arg : p_method_info . arguments ) {
if ( arg . type = = Variant : : INT & & arg . hint = = PROPERTY_HINT_INT_IS_POINTER ) {
return true ;
}
}
return false ;
}
2019-09-24 21:09:54 +08:00
bool BindingsGenerator : : _populate_object_type_interfaces ( ) {
2017-10-03 05:24:00 +08:00
obj_types . clear ( ) ;
List < StringName > class_list ;
ClassDB : : get_class_list ( & class_list ) ;
class_list . sort_custom < StringName : : AlphCompare > ( ) ;
while ( class_list . size ( ) ) {
StringName type_cname = class_list . front ( ) - > get ( ) ;
ClassDB : : APIType api_type = ClassDB : : get_api_type ( type_cname ) ;
if ( api_type = = ClassDB : : API_NONE ) {
class_list . pop_front ( ) ;
continue ;
}
2022-03-19 00:28:48 +08:00
if ( ignored_types . has ( type_cname ) ) {
_log ( " Ignoring type '%s' because it's in the list of ignored types \n " , String ( type_cname ) . utf8 ( ) . get_data ( ) ) ;
class_list . pop_front ( ) ;
continue ;
}
2017-12-24 10:17:48 +08:00
if ( ! ClassDB : : is_class_exposed ( type_cname ) ) {
2019-08-09 09:39:45 +08:00
_log ( " Ignoring type '%s' because it's not exposed \n " , String ( type_cname ) . utf8 ( ) . get_data ( ) ) ;
2017-12-24 10:17:48 +08:00
class_list . pop_front ( ) ;
continue ;
}
2018-07-19 03:15:41 +08:00
if ( ! ClassDB : : is_class_enabled ( type_cname ) ) {
2019-08-09 09:39:45 +08:00
_log ( " Ignoring type '%s' because it's not enabled \n " , String ( type_cname ) . utf8 ( ) . get_data ( ) ) ;
2018-07-19 03:15:41 +08:00
class_list . pop_front ( ) ;
continue ;
}
2017-12-24 10:17:48 +08:00
ClassDB : : ClassInfo * class_info = ClassDB : : classes . getptr ( type_cname ) ;
2022-12-07 23:11:39 +08:00
TypeInterface itype = TypeInterface : : create_object_type ( type_cname , pascal_to_pascal_case ( type_cname ) , api_type ) ;
2017-10-03 05:24:00 +08:00
itype . base_name = ClassDB : : get_parent_class ( type_cname ) ;
2023-02-25 01:02:31 +08:00
itype . is_singleton = Engine : : get_singleton ( ) - > has_singleton ( type_cname ) ;
2019-07-08 21:13:57 +08:00
itype . is_instantiable = class_info - > creation_func & & ! itype . is_singleton ;
2021-06-05 00:03:15 +08:00
itype . is_ref_counted = ClassDB : : is_parent_class ( type_cname , name_cache . type_RefCounted ) ;
itype . memory_own = itype . is_ref_counted ;
2017-10-03 05:24:00 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_out = " %5return " ;
2017-10-03 05:24:00 +08:00
itype . c_out + = C_METHOD_UNMANAGED_GET_MANAGED ;
2021-12-29 06:25:16 +08:00
itype . c_out + = itype . is_ref_counted ? " (%1.Reference); \n " : " (%1); \n " ;
2017-10-03 05:24:00 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . cs_type = itype . proxy_name ;
if ( itype . is_singleton ) {
2022-12-07 23:16:51 +08:00
itype . cs_in_expr = " GodotObject. " CS_STATIC_METHOD_GETINSTANCE " ( " CS_PROPERTY_SINGLETON " ) " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
} else {
2022-12-07 23:16:51 +08:00
itype . cs_in_expr = " GodotObject. " CS_STATIC_METHOD_GETINSTANCE " (%0) " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
}
2017-10-03 05:24:00 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . cs_out = " %5return (%2)%0(%1); " ;
itype . c_arg_in = " (void*)%s " ;
itype . c_type = " IntPtr " ;
2017-10-03 05:24:00 +08:00
itype . c_type_in = itype . c_type ;
2022-12-07 23:16:51 +08:00
itype . c_type_out = " GodotObject " ;
2017-10-03 05:24:00 +08:00
2019-04-25 23:11:01 +08:00
// Populate properties
2017-12-29 09:18:46 +08:00
List < PropertyInfo > property_list ;
ClassDB : : get_property_list ( type_cname , & property_list , true ) ;
2022-05-13 21:04:37 +08:00
HashMap < StringName , StringName > accessor_methods ;
2017-12-29 09:18:46 +08:00
2021-07-16 11:45:57 +08:00
for ( const PropertyInfo & property : property_list ) {
2021-09-08 18:51:45 +08:00
if ( property . usage & PROPERTY_USAGE_GROUP | | property . usage & PROPERTY_USAGE_SUBGROUP | | property . usage & PROPERTY_USAGE_CATEGORY | | ( property . type = = Variant : : NIL & & property . usage & PROPERTY_USAGE_ARRAY ) ) {
2017-12-29 09:18:46 +08:00
continue ;
2020-05-16 10:03:05 +08:00
}
2017-12-29 09:18:46 +08:00
2020-05-10 02:54:07 +08:00
if ( property . name . find ( " / " ) > = 0 ) {
// Ignore properties with '/' (slash) in the name. These are only meant for use in the inspector.
continue ;
}
2017-12-29 09:18:46 +08:00
PropertyInterface iprop ;
iprop . cname = property . name ;
iprop . setter = ClassDB : : get_property_setter ( type_cname , iprop . cname ) ;
iprop . getter = ClassDB : : get_property_getter ( type_cname , iprop . cname ) ;
2020-05-16 10:03:05 +08:00
if ( iprop . setter ! = StringName ( ) ) {
2019-04-25 23:11:01 +08:00
accessor_methods [ iprop . setter ] = iprop . cname ;
2020-05-16 10:03:05 +08:00
}
if ( iprop . getter ! = StringName ( ) ) {
2019-04-25 23:11:01 +08:00
accessor_methods [ iprop . getter ] = iprop . cname ;
2020-05-16 10:03:05 +08:00
}
2019-04-25 23:11:01 +08:00
2017-12-29 09:18:46 +08:00
bool valid = false ;
iprop . index = ClassDB : : get_property_index ( type_cname , iprop . cname , & valid ) ;
2020-05-10 02:54:07 +08:00
ERR_FAIL_COND_V_MSG ( ! valid , false , " Invalid property: ' " + itype . name + " . " + String ( iprop . cname ) + " '. " ) ;
2017-12-29 09:18:46 +08:00
2019-03-19 04:59:35 +08:00
iprop . proxy_name = escape_csharp_keyword ( snake_to_pascal_case ( iprop . cname ) ) ;
2019-04-25 21:45:12 +08:00
// Prevent the property and its enclosing type from sharing the same name
2017-12-29 09:18:46 +08:00
if ( iprop . proxy_name = = itype . proxy_name ) {
2019-08-09 09:39:45 +08:00
_log ( " Name of property '%s' is ambiguous with the name of its enclosing class '%s'. Renaming property to '%s_' \n " ,
2019-04-25 21:45:12 +08:00
iprop . proxy_name . utf8 ( ) . get_data ( ) , itype . proxy_name . utf8 ( ) . get_data ( ) , iprop . proxy_name . utf8 ( ) . get_data ( ) ) ;
2017-12-29 09:18:46 +08:00
iprop . proxy_name + = " _ " ;
}
2020-04-02 07:20:12 +08:00
iprop . prop_doc = nullptr ;
2017-12-29 09:18:46 +08:00
for ( int i = 0 ; i < itype . class_doc - > properties . size ( ) ; i + + ) {
const DocData : : PropertyDoc & prop_doc = itype . class_doc - > properties [ i ] ;
if ( prop_doc . name = = iprop . cname ) {
iprop . prop_doc = & prop_doc ;
break ;
}
}
itype . properties . push_back ( iprop ) ;
}
2017-12-24 10:17:48 +08:00
// Populate methods
2017-10-03 05:24:00 +08:00
List < MethodInfo > virtual_method_list ;
ClassDB : : get_virtual_methods ( type_cname , & virtual_method_list , true ) ;
List < MethodInfo > method_list ;
ClassDB : : get_method_list ( type_cname , & method_list , true ) ;
method_list . sort ( ) ;
2021-07-26 00:45:00 +08:00
for ( const MethodInfo & method_info : method_list ) {
2017-10-03 05:24:00 +08:00
int argc = method_info . arguments . size ( ) ;
2020-12-15 20:04:21 +08:00
if ( method_info . name . is_empty ( ) ) {
2017-10-03 05:24:00 +08:00
continue ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
2019-05-24 06:40:16 +08:00
String cname = method_info . name ;
2020-05-16 10:03:05 +08:00
if ( blacklisted_methods . find ( itype . cname ) & & blacklisted_methods [ itype . cname ] . find ( cname ) ) {
2019-05-24 06:40:16 +08:00
continue ;
2020-05-16 10:03:05 +08:00
}
2019-05-24 06:40:16 +08:00
2023-01-17 11:32:05 +08:00
if ( method_has_ptr_parameter ( method_info ) ) {
// Pointers are not supported.
2023-03-24 18:40:54 +08:00
itype . ignored_members . insert ( method_info . name ) ;
2023-01-17 11:32:05 +08:00
continue ;
}
2017-10-03 05:24:00 +08:00
MethodInterface imethod ;
imethod . name = method_info . name ;
2019-05-24 06:40:16 +08:00
imethod . cname = cname ;
2017-10-03 05:24:00 +08:00
2022-03-21 08:25:50 +08:00
if ( method_info . flags & METHOD_FLAG_STATIC ) {
imethod . is_static = true ;
}
2020-05-16 10:03:05 +08:00
if ( method_info . flags & METHOD_FLAG_VIRTUAL ) {
2017-10-03 05:24:00 +08:00
imethod . is_virtual = true ;
2021-09-13 02:21:15 +08:00
itype . has_virtual_methods = true ;
2020-05-16 10:03:05 +08:00
}
2017-10-03 05:24:00 +08:00
PropertyInfo return_info = method_info . return_val ;
2020-04-02 07:20:12 +08:00
MethodBind * m = imethod . is_virtual ? nullptr : ClassDB : : get_method ( type_cname , method_info . name ) ;
2017-10-03 05:24:00 +08:00
imethod . is_vararg = m & & m - > is_vararg ( ) ;
if ( ! m & & ! imethod . is_virtual ) {
2019-09-24 21:09:54 +08:00
ERR_FAIL_COND_V_MSG ( ! virtual_method_list . find ( method_info ) , false ,
2019-08-09 09:39:45 +08:00
" Missing MethodBind for non-virtual method: ' " + itype . name + " . " + imethod . name + " '. " ) ;
// A virtual method without the virtual flag. This is a special case.
// There is no method bind, so let's fallback to Godot's object.Call(string, params)
imethod . requires_object_call = true ;
// The method Object.free is registered as a virtual method, but without the virtual flag.
// This is because this method is not supposed to be overridden, but called.
// We assume the return type is void.
imethod . return_type . cname = name_cache . type_void ;
2020-05-16 10:03:05 +08:00
// Actually, more methods like this may be added in the future, which could return
// something different. Let's put this check to notify us if that ever happens.
2019-08-09 09:39:45 +08:00
if ( itype . cname ! = name_cache . type_Object | | imethod . name ! = " free " ) {
2019-11-07 16:44:15 +08:00
WARN_PRINT ( " Notification: New unexpected virtual non-overridable method found. "
" We only expected Object.free, but found ' " +
2021-10-28 21:19:35 +08:00
itype . name + " . " + imethod . name + " '. " ) ;
2017-10-03 05:24:00 +08:00
}
2022-06-24 17:16:37 +08:00
} else if ( return_info . type = = Variant : : INT & & return_info . usage & ( PROPERTY_USAGE_CLASS_IS_ENUM | PROPERTY_USAGE_CLASS_IS_BITFIELD ) ) {
2018-04-29 04:25:25 +08:00
imethod . return_type . cname = return_info . class_name ;
imethod . return_type . is_enum = true ;
2017-10-03 05:24:00 +08:00
} else if ( return_info . class_name ! = StringName ( ) ) {
2018-04-29 04:25:25 +08:00
imethod . return_type . cname = return_info . class_name ;
2020-05-16 10:03:05 +08:00
bool bad_reference_hint = ! imethod . is_virtual & & return_info . hint ! = PROPERTY_HINT_RESOURCE_TYPE & &
2021-10-28 21:19:35 +08:00
ClassDB : : is_parent_class ( return_info . class_name , name_cache . type_RefCounted ) ;
2020-05-16 10:03:05 +08:00
ERR_FAIL_COND_V_MSG ( bad_reference_hint , false ,
String ( ) + " Return type is reference but hint is not ' " _STR ( PROPERTY_HINT_RESOURCE_TYPE ) " '. " +
" Are you returning a reference type by pointer? Method: ' " + itype . name + " . " + imethod . name + " '. " ) ;
2022-02-09 07:56:10 +08:00
} else if ( return_info . type = = Variant : : ARRAY & & return_info . hint = = PROPERTY_HINT_ARRAY_TYPE ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
imethod . return_type . cname = Variant : : get_type_name ( return_info . type ) + " _@generic " ;
2022-02-09 07:56:10 +08:00
imethod . return_type . generic_type_parameters . push_back ( TypeReference ( return_info . hint_string ) ) ;
2017-10-03 05:24:00 +08:00
} else if ( return_info . hint = = PROPERTY_HINT_RESOURCE_TYPE ) {
2018-04-29 04:25:25 +08:00
imethod . return_type . cname = return_info . hint_string ;
2017-10-03 05:24:00 +08:00
} else if ( return_info . type = = Variant : : NIL & & return_info . usage & PROPERTY_USAGE_NIL_IS_VARIANT ) {
2018-04-29 04:25:25 +08:00
imethod . return_type . cname = name_cache . type_Variant ;
2017-10-03 05:24:00 +08:00
} else if ( return_info . type = = Variant : : NIL ) {
2018-04-29 04:25:25 +08:00
imethod . return_type . cname = name_cache . type_void ;
2017-10-03 05:24:00 +08:00
} else {
2023-01-25 10:51:32 +08:00
imethod . return_type . cname = _get_type_name_from_meta ( return_info . type , m ? m - > get_argument_meta ( - 1 ) : ( GodotTypeInfo : : Metadata ) method_info . return_val_metadata ) ;
2017-10-03 05:24:00 +08:00
}
for ( int i = 0 ; i < argc ; i + + ) {
PropertyInfo arginfo = method_info . arguments [ i ] ;
2019-09-24 21:09:54 +08:00
String orig_arg_name = arginfo . name ;
2017-10-03 05:24:00 +08:00
ArgumentInterface iarg ;
2019-09-24 21:09:54 +08:00
iarg . name = orig_arg_name ;
2017-10-03 05:24:00 +08:00
2022-06-24 17:16:37 +08:00
if ( arginfo . type = = Variant : : INT & & arginfo . usage & ( PROPERTY_USAGE_CLASS_IS_ENUM | PROPERTY_USAGE_CLASS_IS_BITFIELD ) ) {
2018-04-29 04:25:25 +08:00
iarg . type . cname = arginfo . class_name ;
iarg . type . is_enum = true ;
2017-10-03 05:24:00 +08:00
} else if ( arginfo . class_name ! = StringName ( ) ) {
2018-04-29 04:25:25 +08:00
iarg . type . cname = arginfo . class_name ;
2022-02-09 07:56:10 +08:00
} else if ( arginfo . type = = Variant : : ARRAY & & arginfo . hint = = PROPERTY_HINT_ARRAY_TYPE ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
iarg . type . cname = Variant : : get_type_name ( arginfo . type ) + " _@generic " ;
2022-02-09 07:56:10 +08:00
iarg . type . generic_type_parameters . push_back ( TypeReference ( arginfo . hint_string ) ) ;
2017-10-03 05:24:00 +08:00
} else if ( arginfo . hint = = PROPERTY_HINT_RESOURCE_TYPE ) {
2018-04-29 04:25:25 +08:00
iarg . type . cname = arginfo . hint_string ;
2017-10-03 05:24:00 +08:00
} else if ( arginfo . type = = Variant : : NIL ) {
2018-04-29 04:25:25 +08:00
iarg . type . cname = name_cache . type_Variant ;
2017-10-03 05:24:00 +08:00
} else {
2023-01-25 10:51:32 +08:00
iarg . type . cname = _get_type_name_from_meta ( arginfo . type , m ? m - > get_argument_meta ( i ) : ( GodotTypeInfo : : Metadata ) method_info . get_argument_meta ( i ) ) ;
2017-10-03 05:24:00 +08:00
}
iarg . name = escape_csharp_keyword ( snake_to_camel_case ( iarg . name ) ) ;
if ( m & & m - > has_default_argument ( i ) ) {
2019-09-24 21:09:54 +08:00
bool defval_ok = _arg_default_value_from_variant ( m - > get_default_argument ( i ) , iarg ) ;
ERR_FAIL_COND_V_MSG ( ! defval_ok , false ,
" Cannot determine default value for argument ' " + orig_arg_name + " ' of method ' " + itype . name + " . " + imethod . name + " '. " ) ;
2017-10-03 05:24:00 +08:00
}
imethod . add_argument ( iarg ) ;
}
if ( imethod . is_vararg ) {
ArgumentInterface ivararg ;
2018-04-29 04:25:25 +08:00
ivararg . type . cname = name_cache . type_VarArg ;
2017-10-03 05:24:00 +08:00
ivararg . name = " @args " ;
imethod . add_argument ( ivararg ) ;
}
imethod . proxy_name = escape_csharp_keyword ( snake_to_pascal_case ( imethod . name ) ) ;
2019-04-25 21:45:12 +08:00
// Prevent the method and its enclosing type from sharing the same name
2017-10-03 05:24:00 +08:00
if ( imethod . proxy_name = = itype . proxy_name ) {
2019-08-09 09:39:45 +08:00
_log ( " Name of method '%s' is ambiguous with the name of its enclosing class '%s'. Renaming method to '%s_' \n " ,
2019-04-25 21:45:12 +08:00
imethod . proxy_name . utf8 ( ) . get_data ( ) , itype . proxy_name . utf8 ( ) . get_data ( ) , imethod . proxy_name . utf8 ( ) . get_data ( ) ) ;
2017-10-03 05:24:00 +08:00
imethod . proxy_name + = " _ " ;
}
2022-05-13 21:04:37 +08:00
HashMap < StringName , StringName > : : Iterator accessor = accessor_methods . find ( imethod . cname ) ;
2019-04-25 23:11:01 +08:00
if ( accessor ) {
2023-01-16 21:57:50 +08:00
// We only make internal an accessor method if it's in the same class as the property.
// It's easier this way, but also we don't know if an accessor method in a different class
// could have other purposes, so better leave those untouched.
imethod . is_internal = true ;
2019-04-25 23:11:01 +08:00
}
2017-10-03 05:24:00 +08:00
if ( itype . class_doc ) {
for ( int i = 0 ; i < itype . class_doc - > methods . size ( ) ; i + + ) {
if ( itype . class_doc - > methods [ i ] . name = = imethod . name ) {
imethod . method_doc = & itype . class_doc - > methods [ i ] ;
break ;
}
}
}
2020-05-16 10:03:05 +08:00
ERR_FAIL_COND_V_MSG ( itype . find_property_by_name ( imethod . cname ) , false ,
" Method name conflicts with property: ' " + itype . name + " . " + imethod . name + " '. " ) ;
2021-09-13 02:21:15 +08:00
// Methods starting with an underscore are ignored unless they're used as a property setter or getter
2017-10-03 05:24:00 +08:00
if ( ! imethod . is_virtual & & imethod . name [ 0 ] = = ' _ ' ) {
2021-07-26 00:45:00 +08:00
for ( const PropertyInterface & iprop : itype . properties ) {
2017-12-29 09:18:46 +08:00
if ( iprop . setter = = imethod . name | | iprop . getter = = imethod . name ) {
2017-10-03 05:24:00 +08:00
imethod . is_internal = true ;
itype . methods . push_back ( imethod ) ;
break ;
}
}
} else {
itype . methods . push_back ( imethod ) ;
}
}
2020-03-15 02:20:17 +08:00
// Populate signals
const HashMap < StringName , MethodInfo > & signal_map = class_info - > signal_map ;
2022-05-08 16:09:19 +08:00
for ( const KeyValue < StringName , MethodInfo > & E : signal_map ) {
2020-03-15 02:20:17 +08:00
SignalInterface isignal ;
2022-05-08 16:09:19 +08:00
const MethodInfo & method_info = E . value ;
2020-03-15 02:20:17 +08:00
isignal . name = method_info . name ;
isignal . cname = method_info . name ;
int argc = method_info . arguments . size ( ) ;
for ( int i = 0 ; i < argc ; i + + ) {
PropertyInfo arginfo = method_info . arguments [ i ] ;
String orig_arg_name = arginfo . name ;
ArgumentInterface iarg ;
iarg . name = orig_arg_name ;
2022-06-24 17:16:37 +08:00
if ( arginfo . type = = Variant : : INT & & arginfo . usage & ( PROPERTY_USAGE_CLASS_IS_ENUM | PROPERTY_USAGE_CLASS_IS_BITFIELD ) ) {
2020-03-15 02:20:17 +08:00
iarg . type . cname = arginfo . class_name ;
iarg . type . is_enum = true ;
} else if ( arginfo . class_name ! = StringName ( ) ) {
iarg . type . cname = arginfo . class_name ;
2022-02-09 07:56:10 +08:00
} else if ( arginfo . type = = Variant : : ARRAY & & arginfo . hint = = PROPERTY_HINT_ARRAY_TYPE ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
iarg . type . cname = Variant : : get_type_name ( arginfo . type ) + " _@generic " ;
2022-02-09 07:56:10 +08:00
iarg . type . generic_type_parameters . push_back ( TypeReference ( arginfo . hint_string ) ) ;
2020-03-15 02:20:17 +08:00
} else if ( arginfo . hint = = PROPERTY_HINT_RESOURCE_TYPE ) {
iarg . type . cname = arginfo . hint_string ;
} else if ( arginfo . type = = Variant : : NIL ) {
iarg . type . cname = name_cache . type_Variant ;
} else {
2023-01-25 10:51:32 +08:00
iarg . type . cname = _get_type_name_from_meta ( arginfo . type , ( GodotTypeInfo : : Metadata ) method_info . get_argument_meta ( i ) ) ;
2020-03-15 02:20:17 +08:00
}
iarg . name = escape_csharp_keyword ( snake_to_camel_case ( iarg . name ) ) ;
isignal . add_argument ( iarg ) ;
}
isignal . proxy_name = escape_csharp_keyword ( snake_to_pascal_case ( isignal . name ) ) ;
// Prevent the signal and its enclosing type from sharing the same name
if ( isignal . proxy_name = = itype . proxy_name ) {
_log ( " Name of signal '%s' is ambiguous with the name of its enclosing class '%s'. Renaming signal to '%s_' \n " ,
isignal . proxy_name . utf8 ( ) . get_data ( ) , itype . proxy_name . utf8 ( ) . get_data ( ) , isignal . proxy_name . utf8 ( ) . get_data ( ) ) ;
isignal . proxy_name + = " _ " ;
}
if ( itype . find_property_by_proxy_name ( isignal . proxy_name ) | | itype . find_method_by_proxy_name ( isignal . proxy_name ) ) {
// ClassDB allows signal names that conflict with method or property names.
// While registering a signal with a conflicting name is considered wrong,
// it may still happen and it may take some time until someone fixes the name.
// We can't allow the bindings to be in a broken state while we wait for a fix;
// that's why we must handle this possibility by renaming the signal.
isignal . proxy_name + = " Signal " ;
}
if ( itype . class_doc ) {
for ( int i = 0 ; i < itype . class_doc - > signals . size ( ) ; i + + ) {
const DocData : : MethodDoc & signal_doc = itype . class_doc - > signals [ i ] ;
if ( signal_doc . name = = isignal . name ) {
isignal . method_doc = & signal_doc ;
break ;
}
}
}
itype . signals_ . push_back ( isignal ) ;
}
2017-12-24 10:17:48 +08:00
// Populate enums and constants
2019-04-19 07:10:08 +08:00
List < String > constants ;
ClassDB : : get_integer_constant_list ( type_cname , & constants , true ) ;
2017-12-24 10:17:48 +08:00
2022-06-24 17:16:37 +08:00
const HashMap < StringName , ClassDB : : ClassInfo : : EnumInfo > & enum_map = class_info - > enum_map ;
2017-12-24 10:17:48 +08:00
2022-06-24 17:16:37 +08:00
for ( const KeyValue < StringName , ClassDB : : ClassInfo : : EnumInfo > & E : enum_map ) {
2022-05-08 16:09:19 +08:00
StringName enum_proxy_cname = E . key ;
2022-12-07 23:11:39 +08:00
String enum_proxy_name = pascal_to_pascal_case ( enum_proxy_cname . operator String ( ) ) ;
2022-10-15 18:47:10 +08:00
if ( itype . find_property_by_proxy_name ( enum_proxy_name ) | | itype . find_method_by_proxy_name ( enum_proxy_name ) | | itype . find_signal_by_proxy_name ( enum_proxy_name ) ) {
// In case the enum name conflicts with other PascalCase members,
// we append 'Enum' to the enum name in those cases.
// We have several conflicts between enums and PascalCase properties.
2017-12-24 10:17:48 +08:00
enum_proxy_name + = " Enum " ;
enum_proxy_cname = StringName ( enum_proxy_name ) ;
}
EnumInterface ienum ( enum_proxy_cname ) ;
2022-06-25 02:25:30 +08:00
ienum . is_flags = E . value . is_bitfield ;
2022-06-24 17:16:37 +08:00
const List < StringName > & enum_constants = E . value . constants ;
2021-07-26 00:45:00 +08:00
for ( const StringName & constant_cname : enum_constants ) {
2018-10-19 01:41:10 +08:00
String constant_name = constant_cname . operator String ( ) ;
2022-05-09 17:47:10 +08:00
int64_t * value = class_info - > constant_map . getptr ( constant_cname ) ;
2019-09-24 21:09:54 +08:00
ERR_FAIL_NULL_V ( value , false ) ;
2019-04-19 07:10:08 +08:00
constants . erase ( constant_name ) ;
2017-12-24 10:17:48 +08:00
2018-10-19 01:41:10 +08:00
ConstantInterface iconstant ( constant_name , snake_to_pascal_case ( constant_name , true ) , * value ) ;
2017-12-24 10:17:48 +08:00
2020-04-02 07:20:12 +08:00
iconstant . const_doc = nullptr ;
2017-12-24 10:17:48 +08:00
for ( int i = 0 ; i < itype . class_doc - > constants . size ( ) ; i + + ) {
const DocData : : ConstantDoc & const_doc = itype . class_doc - > constants [ i ] ;
if ( const_doc . name = = iconstant . name ) {
iconstant . const_doc = & const_doc ;
break ;
}
}
ienum . constants . push_back ( iconstant ) ;
}
2018-10-19 01:41:10 +08:00
int prefix_length = _determine_enum_prefix ( ienum ) ;
_apply_prefix_to_enum_constants ( ienum , prefix_length ) ;
2017-12-24 10:17:48 +08:00
itype . enums . push_back ( ienum ) ;
TypeInterface enum_itype ;
2018-04-29 04:25:25 +08:00
enum_itype . is_enum = true ;
2022-05-08 16:09:19 +08:00
enum_itype . name = itype . name + " . " + String ( E . key ) ;
2017-12-24 10:17:48 +08:00
enum_itype . cname = StringName ( enum_itype . name ) ;
enum_itype . proxy_name = itype . proxy_name + " . " + enum_proxy_name ;
2018-04-29 04:25:25 +08:00
TypeInterface : : postsetup_enum_type ( enum_itype ) ;
2017-12-24 10:17:48 +08:00
enum_types . insert ( enum_itype . cname , enum_itype ) ;
}
2021-07-26 00:45:00 +08:00
for ( const String & constant_name : constants ) {
2022-05-09 17:47:10 +08:00
int64_t * value = class_info - > constant_map . getptr ( StringName ( constant_name ) ) ;
2019-09-24 21:09:54 +08:00
ERR_FAIL_NULL_V ( value , false ) ;
2017-12-24 10:17:48 +08:00
2022-10-15 18:47:10 +08:00
String constant_proxy_name = snake_to_pascal_case ( constant_name , true ) ;
if ( itype . find_property_by_proxy_name ( constant_proxy_name ) | | itype . find_method_by_proxy_name ( constant_proxy_name ) | | itype . find_signal_by_proxy_name ( constant_proxy_name ) ) {
// In case the constant name conflicts with other PascalCase members,
// we append 'Constant' to the constant name in those cases.
constant_proxy_name + = " Constant " ;
}
ConstantInterface iconstant ( constant_name , constant_proxy_name , * value ) ;
2017-12-24 10:17:48 +08:00
2020-04-02 07:20:12 +08:00
iconstant . const_doc = nullptr ;
2017-12-24 10:17:48 +08:00
for ( int i = 0 ; i < itype . class_doc - > constants . size ( ) ; i + + ) {
const DocData : : ConstantDoc & const_doc = itype . class_doc - > constants [ i ] ;
if ( const_doc . name = = iconstant . name ) {
iconstant . const_doc = & const_doc ;
break ;
}
}
itype . constants . push_back ( iconstant ) ;
}
obj_types . insert ( itype . cname , itype ) ;
2017-10-03 05:24:00 +08:00
class_list . pop_front ( ) ;
}
2019-09-24 21:09:54 +08:00
return true ;
2017-10-03 05:24:00 +08:00
}
2019-09-24 21:09:54 +08:00
bool BindingsGenerator : : _arg_default_value_from_variant ( const Variant & p_val , ArgumentInterface & r_iarg ) {
2020-05-16 10:03:05 +08:00
r_iarg . def_param_value = p_val ;
r_iarg . default_argument = p_val . operator String ( ) ;
2017-10-03 05:24:00 +08:00
switch ( p_val . get_type ( ) ) {
case Variant : : NIL :
2019-06-15 23:34:07 +08:00
// Either Object type or Variant
2022-07-28 23:41:49 +08:00
r_iarg . default_argument = " default " ;
2017-10-03 05:24:00 +08:00
break ;
// Atomic types
case Variant : : BOOL :
r_iarg . default_argument = bool ( p_val ) ? " true " : " false " ;
break ;
case Variant : : INT :
2018-04-29 04:25:25 +08:00
if ( r_iarg . type . cname ! = name_cache . type_int ) {
2017-12-24 10:17:48 +08:00
r_iarg . default_argument = " (%s) " + r_iarg . default_argument ;
}
break ;
Variant: Added 64-bit packed arrays, renamed Variant::REAL to FLOAT.
- Renames PackedIntArray to PackedInt32Array.
- Renames PackedFloatArray to PackedFloat32Array.
- Adds PackedInt64Array and PackedFloat64Array.
- Renames Variant::REAL to Variant::FLOAT for consistency.
Packed arrays are for storing large amount of data and creating stuff like
meshes, buffers. textures, etc. Forcing them to be 64 is a huge waste of
memory. That said, many users requested the ability to have 64 bits packed
arrays for their games, so this is just an optional added type.
For Variant, the float datatype is always 64 bits, and exposed as `float`.
We still have `real_t` which is the datatype that can change from 32 to 64
bits depending on a compile flag (not entirely working right now, but that's
the idea). It affects math related datatypes and code only.
Neither Variant nor PackedArray make use of real_t, which is only intended
for math precision, so the term is removed from there to keep only float.
2020-02-25 02:20:53 +08:00
case Variant : : FLOAT :
2020-07-24 15:48:21 +08:00
if ( r_iarg . type . cname = = name_cache . type_float ) {
r_iarg . default_argument + = " f " ;
}
2017-10-03 05:24:00 +08:00
break ;
case Variant : : STRING :
2020-03-15 02:20:17 +08:00
case Variant : : STRING_NAME :
2017-10-03 05:24:00 +08:00
case Variant : : NODE_PATH :
2020-03-15 02:20:17 +08:00
if ( r_iarg . type . cname = = name_cache . type_StringName | | r_iarg . type . cname = = name_cache . type_NodePath ) {
2021-12-29 06:25:16 +08:00
if ( r_iarg . default_argument . length ( ) > 0 ) {
r_iarg . default_argument = " (%s) \" " + r_iarg . default_argument + " \" " ;
r_iarg . def_param_mode = ArgumentInterface : : NULLABLE_REF ;
} else {
// No need for a special `in` statement to change `null` to `""`. Marshaling takes care of this already.
r_iarg . default_argument = " null " ;
}
2020-03-15 02:20:17 +08:00
} else {
CRASH_COND ( r_iarg . type . cname ! = name_cache . type_String ) ;
r_iarg . default_argument = " \" " + r_iarg . default_argument + " \" " ;
}
2017-10-03 05:24:00 +08:00
break ;
2020-07-24 15:48:21 +08:00
case Variant : : PLANE : {
Plane plane = p_val . operator Plane ( ) ;
2021-02-25 22:54:50 +08:00
r_iarg . default_argument = " new Plane(new Vector3 " + plane . normal . operator String ( ) + " , " + rtos ( plane . d ) + " ) " ;
2017-10-03 05:24:00 +08:00
r_iarg . def_param_mode = ArgumentInterface : : NULLABLE_VAL ;
2020-07-24 15:48:21 +08:00
} break ;
case Variant : : AABB : {
AABB aabb = p_val . operator : : AABB ( ) ;
2022-12-07 23:11:39 +08:00
r_iarg . default_argument = " new Aabb(new Vector3 " + aabb . position . operator String ( ) + " , new Vector3 " + aabb . size . operator String ( ) + " ) " ;
2020-07-24 15:48:21 +08:00
r_iarg . def_param_mode = ArgumentInterface : : NULLABLE_VAL ;
} break ;
case Variant : : RECT2 : {
Rect2 rect = p_val . operator Rect2 ( ) ;
2021-06-18 13:13:00 +08:00
r_iarg . default_argument = " new Rect2(new Vector2 " + rect . position . operator String ( ) + " , new Vector2 " + rect . size . operator String ( ) + " ) " ;
2020-07-24 15:48:21 +08:00
r_iarg . def_param_mode = ArgumentInterface : : NULLABLE_VAL ;
} break ;
case Variant : : RECT2I : {
Rect2i rect = p_val . operator Rect2i ( ) ;
2022-12-07 23:11:39 +08:00
r_iarg . default_argument = " new Rect2I(new Vector2I " + rect . position . operator String ( ) + " , new Vector2I " + rect . size . operator String ( ) + " ) " ;
2020-07-24 15:48:21 +08:00
r_iarg . def_param_mode = ArgumentInterface : : NULLABLE_VAL ;
} break ;
2017-10-03 05:24:00 +08:00
case Variant : : COLOR :
case Variant : : VECTOR2 :
2020-03-03 17:42:20 +08:00
case Variant : : VECTOR2I :
2017-10-03 05:24:00 +08:00
case Variant : : VECTOR3 :
2020-03-03 17:42:20 +08:00
case Variant : : VECTOR3I :
2017-10-03 05:24:00 +08:00
r_iarg . default_argument = " new %s " + r_iarg . default_argument ;
r_iarg . def_param_mode = ArgumentInterface : : NULLABLE_VAL ;
break ;
Implement Vector4, Vector4i, Projection
Implement built-in classes Vector4, Vector4i and Projection.
* Two versions of Vector4 (float and integer).
* A Projection class, which is a 4x4 matrix specialized in projection types.
These types have been requested for a long time, but given they were very corner case they were not added before.
Because in Godot 4, reimplementing parts of the rendering engine is now possible, access to these types (heavily used by the rendering code) becomes a necessity.
**Q**: Why Projection and not Matrix4?
**A**: Godot does not use Matrix2, Matrix3, Matrix4x3, etc. naming convention because, within the engine, these types always have a *purpose*. As such, Godot names them: Transform2D, Transform3D or Basis. In this case, this 4x4 matrix is _always_ used as a _Projection_, hence the naming.
2022-07-20 07:11:13 +08:00
case Variant : : VECTOR4 :
case Variant : : VECTOR4I :
r_iarg . default_argument = " new %s " + r_iarg . default_argument ;
r_iarg . def_param_mode = ArgumentInterface : : NULLABLE_VAL ;
break ;
2017-10-03 05:24:00 +08:00
case Variant : : OBJECT :
2019-09-24 21:09:54 +08:00
ERR_FAIL_COND_V_MSG ( ! p_val . is_zero ( ) , false ,
" Parameter of type ' " + String ( r_iarg . type . cname ) + " ' can only have null/zero as the default value. " ) ;
r_iarg . default_argument = " null " ;
break ;
2017-10-03 05:24:00 +08:00
case Variant : : DICTIONARY :
2021-12-29 06:25:16 +08:00
ERR_FAIL_COND_V_MSG ( ! p_val . operator Dictionary ( ) . is_empty ( ) , false ,
" Default value of type 'Dictionary' must be an empty dictionary. " ) ;
// The [cs_in] expression already interprets null values as empty dictionaries.
r_iarg . default_argument = " null " ;
r_iarg . def_param_mode = ArgumentInterface : : CONSTANT ;
2017-10-03 05:24:00 +08:00
break ;
2020-11-09 21:53:05 +08:00
case Variant : : RID :
2019-09-24 21:09:54 +08:00
ERR_FAIL_COND_V_MSG ( r_iarg . type . cname ! = name_cache . type_RID , false ,
" Parameter of type ' " + String ( r_iarg . type . cname ) + " ' cannot have a default value of type ' " + String ( name_cache . type_RID ) + " '. " ) ;
ERR_FAIL_COND_V_MSG ( ! p_val . is_zero ( ) , false ,
" Parameter of type ' " + String ( r_iarg . type . cname ) + " ' can only have null/zero as the default value. " ) ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
r_iarg . default_argument = " default " ;
2019-09-24 21:09:54 +08:00
break ;
2017-10-03 05:24:00 +08:00
case Variant : : ARRAY :
2021-12-29 06:25:16 +08:00
ERR_FAIL_COND_V_MSG ( ! p_val . operator Array ( ) . is_empty ( ) , false ,
" Default value of type 'Array' must be an empty array. " ) ;
// The [cs_in] expression already interprets null values as empty arrays.
r_iarg . default_argument = " null " ;
r_iarg . def_param_mode = ArgumentInterface : : CONSTANT ;
2021-07-25 19:46:31 +08:00
break ;
2020-02-18 05:06:54 +08:00
case Variant : : PACKED_BYTE_ARRAY :
Variant: Added 64-bit packed arrays, renamed Variant::REAL to FLOAT.
- Renames PackedIntArray to PackedInt32Array.
- Renames PackedFloatArray to PackedFloat32Array.
- Adds PackedInt64Array and PackedFloat64Array.
- Renames Variant::REAL to Variant::FLOAT for consistency.
Packed arrays are for storing large amount of data and creating stuff like
meshes, buffers. textures, etc. Forcing them to be 64 is a huge waste of
memory. That said, many users requested the ability to have 64 bits packed
arrays for their games, so this is just an optional added type.
For Variant, the float datatype is always 64 bits, and exposed as `float`.
We still have `real_t` which is the datatype that can change from 32 to 64
bits depending on a compile flag (not entirely working right now, but that's
the idea). It affects math related datatypes and code only.
Neither Variant nor PackedArray make use of real_t, which is only intended
for math precision, so the term is removed from there to keep only float.
2020-02-25 02:20:53 +08:00
case Variant : : PACKED_INT32_ARRAY :
case Variant : : PACKED_INT64_ARRAY :
2020-03-15 02:20:17 +08:00
case Variant : : PACKED_FLOAT32_ARRAY :
Variant: Added 64-bit packed arrays, renamed Variant::REAL to FLOAT.
- Renames PackedIntArray to PackedInt32Array.
- Renames PackedFloatArray to PackedFloat32Array.
- Adds PackedInt64Array and PackedFloat64Array.
- Renames Variant::REAL to Variant::FLOAT for consistency.
Packed arrays are for storing large amount of data and creating stuff like
meshes, buffers. textures, etc. Forcing them to be 64 is a huge waste of
memory. That said, many users requested the ability to have 64 bits packed
arrays for their games, so this is just an optional added type.
For Variant, the float datatype is always 64 bits, and exposed as `float`.
We still have `real_t` which is the datatype that can change from 32 to 64
bits depending on a compile flag (not entirely working right now, but that's
the idea). It affects math related datatypes and code only.
Neither Variant nor PackedArray make use of real_t, which is only intended
for math precision, so the term is removed from there to keep only float.
2020-02-25 02:20:53 +08:00
case Variant : : PACKED_FLOAT64_ARRAY :
2020-02-18 05:06:54 +08:00
case Variant : : PACKED_STRING_ARRAY :
case Variant : : PACKED_VECTOR2_ARRAY :
case Variant : : PACKED_VECTOR3_ARRAY :
case Variant : : PACKED_COLOR_ARRAY :
2021-07-25 19:46:31 +08:00
r_iarg . default_argument = " Array.Empty<%s>() " ;
2017-10-03 05:24:00 +08:00
r_iarg . def_param_mode = ArgumentInterface : : NULLABLE_REF ;
break ;
2020-07-24 15:48:21 +08:00
case Variant : : TRANSFORM2D : {
Transform2D transform = p_val . operator Transform2D ( ) ;
if ( transform = = Transform2D ( ) ) {
r_iarg . default_argument = " Transform2D.Identity " ;
} else {
2022-04-25 05:59:24 +08:00
r_iarg . default_argument = " new Transform2D(new Vector2 " + transform . columns [ 0 ] . operator String ( ) + " , new Vector2 " + transform . columns [ 1 ] . operator String ( ) + " , new Vector2 " + transform . columns [ 2 ] . operator String ( ) + " ) " ;
2020-07-24 15:48:21 +08:00
}
2017-10-03 05:24:00 +08:00
r_iarg . def_param_mode = ArgumentInterface : : NULLABLE_VAL ;
2020-07-24 15:48:21 +08:00
} break ;
2021-04-28 15:36:08 +08:00
case Variant : : TRANSFORM3D : {
2020-10-17 13:08:21 +08:00
Transform3D transform = p_val . operator Transform3D ( ) ;
if ( transform = = Transform3D ( ) ) {
2021-02-27 12:26:56 +08:00
r_iarg . default_argument = " Transform3D.Identity " ;
2020-07-24 15:48:21 +08:00
} else {
Basis basis = transform . basis ;
2021-02-27 12:26:56 +08:00
r_iarg . default_argument = " new Transform3D(new Vector3 " + basis . get_column ( 0 ) . operator String ( ) + " , new Vector3 " + basis . get_column ( 1 ) . operator String ( ) + " , new Vector3 " + basis . get_column ( 2 ) . operator String ( ) + " , new Vector3 " + transform . origin . operator String ( ) + " ) " ;
2020-07-24 15:48:21 +08:00
}
Implement Vector4, Vector4i, Projection
Implement built-in classes Vector4, Vector4i and Projection.
* Two versions of Vector4 (float and integer).
* A Projection class, which is a 4x4 matrix specialized in projection types.
These types have been requested for a long time, but given they were very corner case they were not added before.
Because in Godot 4, reimplementing parts of the rendering engine is now possible, access to these types (heavily used by the rendering code) becomes a necessity.
**Q**: Why Projection and not Matrix4?
**A**: Godot does not use Matrix2, Matrix3, Matrix4x3, etc. naming convention because, within the engine, these types always have a *purpose*. As such, Godot names them: Transform2D, Transform3D or Basis. In this case, this 4x4 matrix is _always_ used as a _Projection_, hence the naming.
2022-07-20 07:11:13 +08:00
r_iarg . def_param_mode = ArgumentInterface : : NULLABLE_VAL ;
} break ;
case Variant : : PROJECTION : {
2022-10-05 00:44:48 +08:00
Projection projection = p_val . operator Projection ( ) ;
if ( projection = = Projection ( ) ) {
Implement Vector4, Vector4i, Projection
Implement built-in classes Vector4, Vector4i and Projection.
* Two versions of Vector4 (float and integer).
* A Projection class, which is a 4x4 matrix specialized in projection types.
These types have been requested for a long time, but given they were very corner case they were not added before.
Because in Godot 4, reimplementing parts of the rendering engine is now possible, access to these types (heavily used by the rendering code) becomes a necessity.
**Q**: Why Projection and not Matrix4?
**A**: Godot does not use Matrix2, Matrix3, Matrix4x3, etc. naming convention because, within the engine, these types always have a *purpose*. As such, Godot names them: Transform2D, Transform3D or Basis. In this case, this 4x4 matrix is _always_ used as a _Projection_, hence the naming.
2022-07-20 07:11:13 +08:00
r_iarg . default_argument = " Projection.Identity " ;
} else {
2022-10-05 00:44:48 +08:00
r_iarg . default_argument = " new Projection(new Vector4 " + projection . columns [ 0 ] . operator String ( ) + " , new Vector4 " + projection . columns [ 1 ] . operator String ( ) + " , new Vector4 " + projection . columns [ 2 ] . operator String ( ) + " , new Vector4 " + projection . columns [ 3 ] . operator String ( ) + " ) " ;
Implement Vector4, Vector4i, Projection
Implement built-in classes Vector4, Vector4i and Projection.
* Two versions of Vector4 (float and integer).
* A Projection class, which is a 4x4 matrix specialized in projection types.
These types have been requested for a long time, but given they were very corner case they were not added before.
Because in Godot 4, reimplementing parts of the rendering engine is now possible, access to these types (heavily used by the rendering code) becomes a necessity.
**Q**: Why Projection and not Matrix4?
**A**: Godot does not use Matrix2, Matrix3, Matrix4x3, etc. naming convention because, within the engine, these types always have a *purpose*. As such, Godot names them: Transform2D, Transform3D or Basis. In this case, this 4x4 matrix is _always_ used as a _Projection_, hence the naming.
2022-07-20 07:11:13 +08:00
}
2020-07-24 15:48:21 +08:00
r_iarg . def_param_mode = ArgumentInterface : : NULLABLE_VAL ;
} break ;
case Variant : : BASIS : {
Basis basis = p_val . operator Basis ( ) ;
if ( basis = = Basis ( ) ) {
r_iarg . default_argument = " Basis.Identity " ;
} else {
r_iarg . default_argument = " new Basis(new Vector3 " + basis . get_column ( 0 ) . operator String ( ) + " , new Vector3 " + basis . get_column ( 1 ) . operator String ( ) + " , new Vector3 " + basis . get_column ( 2 ) . operator String ( ) + " ) " ;
}
r_iarg . def_param_mode = ArgumentInterface : : NULLABLE_VAL ;
} break ;
2021-01-20 15:02:02 +08:00
case Variant : : QUATERNION : {
Quaternion quaternion = p_val . operator Quaternion ( ) ;
if ( quaternion = = Quaternion ( ) ) {
r_iarg . default_argument = " Quaternion.Identity " ;
2020-07-24 15:48:21 +08:00
} else {
2021-01-20 15:02:02 +08:00
r_iarg . default_argument = " new Quaternion " + quaternion . operator String ( ) ;
2020-07-24 15:48:21 +08:00
}
r_iarg . def_param_mode = ArgumentInterface : : NULLABLE_VAL ;
} break ;
2020-03-15 02:20:17 +08:00
case Variant : : CALLABLE :
2021-08-20 17:18:04 +08:00
ERR_FAIL_COND_V_MSG ( r_iarg . type . cname ! = name_cache . type_Callable , false ,
" Parameter of type ' " + String ( r_iarg . type . cname ) + " ' cannot have a default value of type ' " + String ( name_cache . type_Callable ) + " '. " ) ;
ERR_FAIL_COND_V_MSG ( ! p_val . is_zero ( ) , false ,
" Parameter of type ' " + String ( r_iarg . type . cname ) + " ' can only have null/zero as the default value. " ) ;
r_iarg . default_argument = " default " ;
break ;
2020-03-15 02:20:17 +08:00
case Variant : : SIGNAL :
2021-08-20 17:18:04 +08:00
ERR_FAIL_COND_V_MSG ( r_iarg . type . cname ! = name_cache . type_Signal , false ,
" Parameter of type ' " + String ( r_iarg . type . cname ) + " ' cannot have a default value of type ' " + String ( name_cache . type_Signal ) + " '. " ) ;
ERR_FAIL_COND_V_MSG ( ! p_val . is_zero ( ) , false ,
" Parameter of type ' " + String ( r_iarg . type . cname ) + " ' can only have null/zero as the default value. " ) ;
r_iarg . default_argument = " default " ;
2020-03-15 02:20:17 +08:00
break ;
default :
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
ERR_FAIL_V_MSG ( false , " Unexpected Variant type: " + itos ( p_val . get_type ( ) ) ) ;
2020-03-15 02:20:17 +08:00
break ;
2017-10-03 05:24:00 +08:00
}
2022-07-28 23:41:49 +08:00
if ( r_iarg . def_param_mode = = ArgumentInterface : : CONSTANT & & r_iarg . type . cname = = name_cache . type_Variant & & r_iarg . default_argument ! = " default " ) {
r_iarg . def_param_mode = ArgumentInterface : : NULLABLE_VAL ;
2020-05-16 10:03:05 +08:00
}
2019-09-24 21:09:54 +08:00
return true ;
2017-10-03 05:24:00 +08:00
}
void BindingsGenerator : : _populate_builtin_type_interfaces ( ) {
builtin_types . clear ( ) ;
TypeInterface itype ;
2022-12-07 23:11:39 +08:00
# define INSERT_STRUCT_TYPE(m_type, m_proxy_name) \
{ \
itype = TypeInterface : : create_value_type ( String ( # m_type ) , String ( # m_proxy_name ) ) ; \
itype . cs_in_expr = " &%0 " ; \
itype . cs_in_expr_is_unsafe = true ; \
builtin_types . insert ( itype . cname , itype ) ; \
}
INSERT_STRUCT_TYPE ( Vector2 , Vector2 )
INSERT_STRUCT_TYPE ( Vector2i , Vector2I )
INSERT_STRUCT_TYPE ( Rect2 , Rect2 )
INSERT_STRUCT_TYPE ( Rect2i , Rect2I )
INSERT_STRUCT_TYPE ( Transform2D , Transform2D )
INSERT_STRUCT_TYPE ( Vector3 , Vector3 )
INSERT_STRUCT_TYPE ( Vector3i , Vector3I )
INSERT_STRUCT_TYPE ( Basis , Basis )
INSERT_STRUCT_TYPE ( Quaternion , Quaternion )
INSERT_STRUCT_TYPE ( Transform3D , Transform3D )
INSERT_STRUCT_TYPE ( AABB , Aabb )
INSERT_STRUCT_TYPE ( Color , Color )
INSERT_STRUCT_TYPE ( Plane , Plane )
INSERT_STRUCT_TYPE ( Vector4 , Vector4 )
INSERT_STRUCT_TYPE ( Vector4i , Vector4I )
INSERT_STRUCT_TYPE ( Projection , Projection )
2017-10-03 05:24:00 +08:00
# undef INSERT_STRUCT_TYPE
2018-01-26 06:44:37 +08:00
// bool
itype = TypeInterface : : create_value_type ( String ( " bool " ) ) ;
2021-12-29 06:25:16 +08:00
itype . cs_in_expr = " %0.ToGodotBool() " ;
2021-09-13 02:23:05 +08:00
itype . cs_out = " %5return %0(%1).ToBool(); " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_type = " godot_bool " ;
itype . c_type_in = itype . c_type ;
itype . c_type_out = itype . c_type ;
itype . c_arg_in = " &%s " ;
itype . c_in_vararg = " %5using godot_variant %1_in = VariantUtils.CreateFromBool(%1); \n " ;
2018-01-26 06:44:37 +08:00
builtin_types . insert ( itype . cname , itype ) ;
2017-10-03 05:24:00 +08:00
2019-04-26 02:24:48 +08:00
// Integer types
2018-10-18 03:37:57 +08:00
{
2019-04-26 02:24:48 +08:00
// C interface for 'uint32_t' is the same as that of enums. Remember to apply
// any of the changes done here to 'TypeInterface::postsetup_enum_type' as well.
2022-02-28 04:57:46 +08:00
# define INSERT_INT_TYPE(m_name, m_int_struct_name) \
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
{ \
itype = TypeInterface : : create_value_type ( String ( m_name ) ) ; \
if ( itype . name ! = " long " & & itype . name ! = " ulong " ) { \
itype . c_in = " %5%0 %1_in = %1; \n " ; \
itype . c_out = " %5return (%0)%1; \n " ; \
itype . c_type = " long " ; \
itype . c_arg_in = " &%s_in " ; \
} else { \
itype . c_arg_in = " &%s " ; \
} \
itype . c_type_in = itype . name ; \
itype . c_type_out = itype . name ; \
itype . c_in_vararg = " %5using godot_variant %1_in = VariantUtils.CreateFromInt(%1); \n " ; \
builtin_types . insert ( itype . cname , itype ) ; \
2018-10-18 03:37:57 +08:00
}
2017-10-03 05:24:00 +08:00
2019-04-26 02:24:48 +08:00
// The expected type for all integers in ptrcall is 'int64_t', so that's what we use for 'c_type'
2022-02-28 04:57:46 +08:00
INSERT_INT_TYPE ( " sbyte " , " Int8 " ) ;
INSERT_INT_TYPE ( " short " , " Int16 " ) ;
INSERT_INT_TYPE ( " int " , " Int32 " ) ;
INSERT_INT_TYPE ( " long " , " Int64 " ) ;
INSERT_INT_TYPE ( " byte " , " UInt8 " ) ;
INSERT_INT_TYPE ( " ushort " , " UInt16 " ) ;
INSERT_INT_TYPE ( " uint " , " UInt32 " ) ;
INSERT_INT_TYPE ( " ulong " , " UInt64 " ) ;
2019-04-26 02:24:48 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
# undef INSERT_INT_TYPE
2022-02-28 04:57:46 +08:00
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
2019-04-26 02:24:48 +08:00
// Floating point types
2018-10-18 03:37:57 +08:00
{
2019-04-26 02:24:48 +08:00
// float
itype = TypeInterface ( ) ;
itype . name = " float " ;
itype . cname = itype . name ;
itype . proxy_name = " float " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . cs_type = itype . proxy_name ;
2019-04-26 02:24:48 +08:00
{
// The expected type for 'float' in ptrcall is 'double'
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_in = " %5%0 %1_in = %1; \n " ;
itype . c_out = " %5return (%0)%1; \n " ;
2019-04-26 02:24:48 +08:00
itype . c_type = " double " ;
itype . c_arg_in = " &%s_in " ;
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_type_in = itype . proxy_name ;
itype . c_type_out = itype . proxy_name ;
itype . c_in_vararg = " %5using godot_variant %1_in = VariantUtils.CreateFromFloat(%1); \n " ;
2019-04-26 02:24:48 +08:00
builtin_types . insert ( itype . cname , itype ) ;
// double
itype = TypeInterface ( ) ;
itype . name = " double " ;
itype . cname = itype . name ;
itype . proxy_name = " double " ;
itype . cs_type = itype . proxy_name ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_type = " double " ;
itype . c_arg_in = " &%s " ;
itype . c_type_in = itype . proxy_name ;
itype . c_type_out = itype . proxy_name ;
itype . c_in_vararg = " %5using godot_variant %1_in = VariantUtils.CreateFromFloat(%1); \n " ;
2019-04-26 02:24:48 +08:00
builtin_types . insert ( itype . cname , itype ) ;
2018-10-18 03:37:57 +08:00
}
2017-10-03 05:24:00 +08:00
// String
itype = TypeInterface ( ) ;
itype . name = " String " ;
2017-12-24 10:17:48 +08:00
itype . cname = itype . name ;
2017-10-03 05:24:00 +08:00
itype . proxy_name = " string " ;
itype . cs_type = itype . proxy_name ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_in = " %5using %0 %1_in = " C_METHOD_MONOSTR_TO_GODOT " (%1); \n " ;
2021-09-13 02:21:15 +08:00
itype . c_out = " %5return " C_METHOD_MONOSTR_FROM_GODOT " (%1); \n " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_arg_in = " &%s_in " ;
itype . c_type = " godot_string " ;
itype . c_type_in = itype . cs_type ;
itype . c_type_out = itype . cs_type ;
itype . c_type_is_disposable_struct = true ;
2022-02-28 04:57:46 +08:00
itype . c_in_vararg = " %5using godot_variant %1_in = VariantUtils.CreateFromString(%1); \n " ;
2017-12-24 10:17:48 +08:00
builtin_types . insert ( itype . cname , itype ) ;
2017-10-03 05:24:00 +08:00
2020-03-15 02:20:17 +08:00
// StringName
itype = TypeInterface ( ) ;
itype . name = " StringName " ;
itype . cname = itype . name ;
itype . proxy_name = " StringName " ;
itype . cs_type = itype . proxy_name ;
2021-12-29 06:25:16 +08:00
itype . cs_in_expr = " (%1)(%0?.NativeValue ?? default) " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
// Cannot pass null StringName to ptrcall
itype . c_out = " %5return %0.CreateTakingOwnershipOfDisposableValue(%1); \n " ;
2021-12-29 06:25:16 +08:00
itype . c_arg_in = " &%s " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_type = " godot_string_name " ;
2021-12-29 06:25:16 +08:00
itype . c_type_in = itype . c_type ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_type_out = itype . cs_type ;
2021-12-29 06:25:16 +08:00
itype . c_in_vararg = " %5using godot_variant %1_in = VariantUtils.CreateFromStringName(%1); \n " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_type_is_disposable_struct = false ; // [c_out] takes ownership
itype . c_ret_needs_default_initialization = true ;
2020-03-15 02:20:17 +08:00
builtin_types . insert ( itype . cname , itype ) ;
2017-10-03 05:24:00 +08:00
// NodePath
itype = TypeInterface ( ) ;
itype . name = " NodePath " ;
2017-12-24 10:17:48 +08:00
itype . cname = itype . name ;
2017-10-03 05:24:00 +08:00
itype . proxy_name = " NodePath " ;
itype . cs_type = itype . proxy_name ;
2021-12-29 06:25:16 +08:00
itype . cs_in_expr = " (%1)(%0?.NativeValue ?? default) " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
// Cannot pass null NodePath to ptrcall
itype . c_out = " %5return %0.CreateTakingOwnershipOfDisposableValue(%1); \n " ;
2021-12-29 06:25:16 +08:00
itype . c_arg_in = " &%s " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_type = " godot_node_path " ;
2021-12-29 06:25:16 +08:00
itype . c_type_in = itype . c_type ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_type_out = itype . cs_type ;
itype . c_type_is_disposable_struct = false ; // [c_out] takes ownership
itype . c_ret_needs_default_initialization = true ;
2017-12-24 10:17:48 +08:00
builtin_types . insert ( itype . cname , itype ) ;
2017-10-03 05:24:00 +08:00
// RID
itype = TypeInterface ( ) ;
itype . name = " RID " ;
2017-12-24 10:17:48 +08:00
itype . cname = itype . name ;
2022-12-07 23:11:39 +08:00
itype . proxy_name = " Rid " ;
2017-10-03 05:24:00 +08:00
itype . cs_type = itype . proxy_name ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_arg_in = " &%s " ;
itype . c_type = itype . cs_type ;
itype . c_type_in = itype . c_type ;
itype . c_type_out = itype . c_type ;
2017-12-24 10:17:48 +08:00
builtin_types . insert ( itype . cname , itype ) ;
2017-10-03 05:24:00 +08:00
// Variant
itype = TypeInterface ( ) ;
itype . name = " Variant " ;
2017-12-24 10:17:48 +08:00
itype . cname = itype . name ;
2022-07-28 23:41:49 +08:00
itype . proxy_name = " Variant " ;
2017-10-03 05:24:00 +08:00
itype . cs_type = itype . proxy_name ;
2022-07-28 23:41:49 +08:00
itype . c_in = " %5%0 %1_in = (%0)%1.NativeVar; \n " ;
itype . c_out = " %5return Variant.CreateTakingOwnershipOfDisposableValue(%1); \n " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_arg_in = " &%s_in " ;
itype . c_type = " godot_variant " ;
itype . c_type_in = itype . cs_type ;
itype . c_type_out = itype . cs_type ;
2022-07-28 23:41:49 +08:00
itype . c_type_is_disposable_struct = false ; // [c_out] takes ownership
itype . c_ret_needs_default_initialization = true ;
2017-12-24 10:17:48 +08:00
builtin_types . insert ( itype . cname , itype ) ;
2017-10-03 05:24:00 +08:00
2020-03-15 02:20:17 +08:00
// Callable
itype = TypeInterface : : create_value_type ( String ( " Callable " ) ) ;
2022-02-28 04:57:46 +08:00
itype . cs_in_expr = " %0 " ;
itype . c_in = " %5using %0 %1_in = " C_METHOD_MANAGED_TO_CALLABLE " (in %1); \n " ;
2021-12-29 06:25:16 +08:00
itype . c_out = " %5return " C_METHOD_MANAGED_FROM_CALLABLE " (in %1); \n " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_arg_in = " &%s_in " ;
itype . c_type = " godot_callable " ;
2022-02-28 04:57:46 +08:00
itype . c_type_in = " in " + itype . cs_type ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_type_out = itype . cs_type ;
itype . c_type_is_disposable_struct = true ;
2020-03-15 02:20:17 +08:00
builtin_types . insert ( itype . cname , itype ) ;
// Signal
itype = TypeInterface ( ) ;
itype . name = " Signal " ;
itype . cname = itype . name ;
2022-12-24 02:28:34 +08:00
itype . proxy_name = " Signal " ;
2020-03-15 02:20:17 +08:00
itype . cs_type = itype . proxy_name ;
2022-02-28 04:57:46 +08:00
itype . cs_in_expr = " %0 " ;
itype . c_in = " %5using %0 %1_in = " C_METHOD_MANAGED_TO_SIGNAL " (in %1); \n " ;
2022-12-10 08:42:36 +08:00
itype . c_out = " %5return " C_METHOD_MANAGED_FROM_SIGNAL " (in %1); \n " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_arg_in = " &%s_in " ;
itype . c_type = " godot_signal " ;
2022-02-28 04:57:46 +08:00
itype . c_type_in = " in " + itype . cs_type ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_type_out = itype . cs_type ;
itype . c_type_is_disposable_struct = true ;
2020-03-15 02:20:17 +08:00
builtin_types . insert ( itype . cname , itype ) ;
2017-10-03 05:24:00 +08:00
// VarArg (fictitious type to represent variable arguments)
itype = TypeInterface ( ) ;
itype . name = " VarArg " ;
2017-12-24 10:17:48 +08:00
itype . cname = itype . name ;
2022-07-28 23:41:49 +08:00
itype . proxy_name = " Variant[] " ;
itype . cs_type = " params Variant[] " ;
itype . cs_in_expr = " %0 ?? Array.Empty<Variant>() " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
// c_type, c_in and c_arg_in are hard-coded in the generator.
// c_out and c_type_out are not applicable to VarArg.
itype . c_arg_in = " &%s_in " ;
2022-07-28 23:41:49 +08:00
itype . c_type_in = " Variant[] " ;
2017-12-24 10:17:48 +08:00
builtin_types . insert ( itype . cname , itype ) ;
2017-10-03 05:24:00 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
# define INSERT_ARRAY_FULL(m_name, m_type, m_managed_type, m_proxy_t) \
{ \
itype = TypeInterface ( ) ; \
itype . name = # m_name ; \
itype . cname = itype . name ; \
itype . proxy_name = # m_proxy_t " [] " ; \
itype . cs_type = itype . proxy_name ; \
itype . c_in = " %5using %0 %1_in = " C_METHOD_MONOARRAY_TO ( m_type ) " (%1); \n " ; \
2021-12-29 06:25:16 +08:00
itype . c_out = " %5return " C_METHOD_MONOARRAY_FROM ( m_type ) " (%1); \n " ; \
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_arg_in = " &%s_in " ; \
itype . c_type = # m_managed_type ; \
itype . c_type_in = itype . proxy_name ; \
itype . c_type_out = itype . proxy_name ; \
itype . c_type_is_disposable_struct = true ; \
builtin_types . insert ( itype . name , itype ) ; \
2017-10-03 05:24:00 +08:00
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
# define INSERT_ARRAY(m_type, m_managed_type, m_proxy_t) INSERT_ARRAY_FULL(m_type, m_type, m_managed_type, m_proxy_t)
2017-10-03 05:24:00 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
INSERT_ARRAY ( PackedInt32Array , godot_packed_int32_array , int ) ;
INSERT_ARRAY ( PackedInt64Array , godot_packed_int64_array , long ) ;
INSERT_ARRAY_FULL ( PackedByteArray , PackedByteArray , godot_packed_byte_array , byte ) ;
2017-10-03 05:24:00 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
INSERT_ARRAY ( PackedFloat32Array , godot_packed_float32_array , float ) ;
INSERT_ARRAY ( PackedFloat64Array , godot_packed_float64_array , double ) ;
2017-10-03 05:24:00 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
INSERT_ARRAY ( PackedStringArray , godot_packed_string_array , string ) ;
2017-10-03 05:24:00 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
INSERT_ARRAY ( PackedColorArray , godot_packed_color_array , Color ) ;
INSERT_ARRAY ( PackedVector2Array , godot_packed_vector2_array , Vector2 ) ;
INSERT_ARRAY ( PackedVector3Array , godot_packed_vector3_array , Vector3 ) ;
2017-10-03 05:24:00 +08:00
# undef INSERT_ARRAY
2018-07-19 05:07:57 +08:00
// Array
itype = TypeInterface ( ) ;
itype . name = " Array " ;
itype . cname = itype . name ;
2018-08-28 02:39:51 +08:00
itype . proxy_name = itype . name ;
2022-02-09 07:56:10 +08:00
itype . type_parameter_count = 1 ;
2018-08-28 02:39:51 +08:00
itype . cs_type = BINDINGS_NAMESPACE_COLLECTIONS " . " + itype . proxy_name ;
2021-12-29 06:25:16 +08:00
itype . cs_in_expr = " (%1)(%0 ?? new()).NativeValue " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_out = " %5return %0.CreateTakingOwnershipOfDisposableValue(%1); \n " ;
2021-12-29 06:25:16 +08:00
itype . c_arg_in = " &%s " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_type = " godot_array " ;
2021-12-29 06:25:16 +08:00
itype . c_type_in = itype . c_type ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_type_out = itype . cs_type ;
itype . c_type_is_disposable_struct = false ; // [c_out] takes ownership
itype . c_ret_needs_default_initialization = true ;
builtin_types . insert ( itype . cname , itype ) ;
// Array_@generic
// Re-use Array's itype
itype . name = " Array_@generic " ;
itype . cname = itype . name ;
itype . cs_out = " %5return new %2(%0(%1)); " ;
2022-12-01 08:45:11 +08:00
// For generic Godot collections, Variant.From<T>/As<T> is slower, so we need this special case
2022-12-24 02:28:34 +08:00
itype . cs_variant_to_managed = " VariantUtils.ConvertToArray(%0) " ;
2022-12-01 08:45:11 +08:00
itype . cs_managed_to_variant = " VariantUtils.CreateFromArray(%0) " ;
2018-07-19 05:07:57 +08:00
builtin_types . insert ( itype . cname , itype ) ;
2017-10-03 05:24:00 +08:00
// Dictionary
itype = TypeInterface ( ) ;
itype . name = " Dictionary " ;
2017-12-24 10:17:48 +08:00
itype . cname = itype . name ;
2018-08-28 02:39:51 +08:00
itype . proxy_name = itype . name ;
2022-02-09 07:56:10 +08:00
itype . type_parameter_count = 2 ;
2018-08-28 02:39:51 +08:00
itype . cs_type = BINDINGS_NAMESPACE_COLLECTIONS " . " + itype . proxy_name ;
2021-12-29 06:25:16 +08:00
itype . cs_in_expr = " (%1)(%0 ?? new()).NativeValue " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_out = " %5return %0.CreateTakingOwnershipOfDisposableValue(%1); \n " ;
2021-12-29 06:25:16 +08:00
itype . c_arg_in = " &%s " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_type = " godot_dictionary " ;
2021-12-29 06:25:16 +08:00
itype . c_type_in = itype . c_type ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . c_type_out = itype . cs_type ;
itype . c_type_is_disposable_struct = false ; // [c_out] takes ownership
itype . c_ret_needs_default_initialization = true ;
builtin_types . insert ( itype . cname , itype ) ;
// Dictionary_@generic
// Re-use Dictionary's itype
itype . name = " Dictionary_@generic " ;
itype . cname = itype . name ;
itype . cs_out = " %5return new %2(%0(%1)); " ;
2022-12-01 08:45:11 +08:00
// For generic Godot collections, Variant.From<T>/As<T> is slower, so we need this special case
2022-12-24 02:28:34 +08:00
itype . cs_variant_to_managed = " VariantUtils.ConvertToDictionary(%0) " ;
2022-12-01 08:45:11 +08:00
itype . cs_managed_to_variant = " VariantUtils.CreateFromDictionary(%0) " ;
2017-12-24 10:17:48 +08:00
builtin_types . insert ( itype . cname , itype ) ;
2017-10-03 05:24:00 +08:00
// void (fictitious type to represent the return type of methods that do not return anything)
itype = TypeInterface ( ) ;
itype . name = " void " ;
2017-12-24 10:17:48 +08:00
itype . cname = itype . name ;
2017-10-03 05:24:00 +08:00
itype . proxy_name = itype . name ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
itype . cs_type = itype . proxy_name ;
itype . c_type = itype . proxy_name ;
2017-10-03 05:24:00 +08:00
itype . c_type_in = itype . c_type ;
itype . c_type_out = itype . c_type ;
2017-12-24 10:17:48 +08:00
builtin_types . insert ( itype . cname , itype ) ;
2017-10-03 05:24:00 +08:00
}
2017-12-24 10:17:48 +08:00
void BindingsGenerator : : _populate_global_constants ( ) {
2020-11-08 06:33:38 +08:00
int global_constants_count = CoreConstants : : get_global_constant_count ( ) ;
2017-12-24 10:17:48 +08:00
if ( global_constants_count > 0 ) {
2022-05-13 21:04:37 +08:00
HashMap < String , DocData : : ClassDoc > : : Iterator match = EditorHelp : : get_doc_data ( ) - > class_list . find ( " @GlobalScope " ) ;
2017-12-24 10:17:48 +08:00
2019-08-09 09:39:45 +08:00
CRASH_COND_MSG ( ! match , " Could not find '@GlobalScope' in DocData. " ) ;
2017-12-24 10:17:48 +08:00
2022-05-13 21:04:37 +08:00
const DocData : : ClassDoc & global_scope_doc = match - > value ;
2017-12-24 10:17:48 +08:00
for ( int i = 0 ; i < global_constants_count ; i + + ) {
2020-11-08 06:33:38 +08:00
String constant_name = CoreConstants : : get_global_constant_name ( i ) ;
2017-12-24 10:17:48 +08:00
2020-04-02 07:20:12 +08:00
const DocData : : ConstantDoc * const_doc = nullptr ;
2018-10-03 06:56:28 +08:00
for ( int j = 0 ; j < global_scope_doc . constants . size ( ) ; j + + ) {
const DocData : : ConstantDoc & curr_const_doc = global_scope_doc . constants [ j ] ;
2017-12-24 10:17:48 +08:00
if ( curr_const_doc . name = = constant_name ) {
const_doc = & curr_const_doc ;
break ;
}
}
2022-05-09 17:47:10 +08:00
int64_t constant_value = CoreConstants : : get_global_constant_value ( i ) ;
2020-11-08 06:33:38 +08:00
StringName enum_name = CoreConstants : : get_global_constant_enum ( i ) ;
2017-12-24 10:17:48 +08:00
2018-10-19 01:41:10 +08:00
ConstantInterface iconstant ( constant_name , snake_to_pascal_case ( constant_name , true ) , constant_value ) ;
2017-12-24 10:17:48 +08:00
iconstant . const_doc = const_doc ;
if ( enum_name ! = StringName ( ) ) {
EnumInterface ienum ( enum_name ) ;
2023-01-09 23:13:28 +08:00
ienum . is_flags = CoreConstants : : is_global_constant_bitfield ( i ) ;
2019-02-13 04:10:08 +08:00
List < EnumInterface > : : Element * enum_match = global_enums . find ( ienum ) ;
if ( enum_match ) {
enum_match - > get ( ) . constants . push_back ( iconstant ) ;
2017-12-24 10:17:48 +08:00
} else {
ienum . constants . push_back ( iconstant ) ;
global_enums . push_back ( ienum ) ;
}
} else {
global_constants . push_back ( iconstant ) ;
}
}
2021-07-26 00:45:00 +08:00
for ( EnumInterface & ienum : global_enums ) {
2017-12-24 10:17:48 +08:00
TypeInterface enum_itype ;
2018-04-29 04:25:25 +08:00
enum_itype . is_enum = true ;
enum_itype . name = ienum . cname . operator String ( ) ;
enum_itype . cname = ienum . cname ;
2022-12-07 23:11:39 +08:00
enum_itype . proxy_name = pascal_to_pascal_case ( enum_itype . name ) ;
2018-04-29 04:25:25 +08:00
TypeInterface : : postsetup_enum_type ( enum_itype ) ;
2017-12-24 10:17:48 +08:00
enum_types . insert ( enum_itype . cname , enum_itype ) ;
2018-10-19 01:41:10 +08:00
int prefix_length = _determine_enum_prefix ( ienum ) ;
2017-12-24 10:17:48 +08:00
2018-10-19 01:41:10 +08:00
// HARDCODED: The Error enum have the prefix 'ERR_' for everything except 'OK' and 'FAILED'.
2017-12-24 10:17:48 +08:00
if ( ienum . cname = = name_cache . enum_Error ) {
2018-10-19 01:41:10 +08:00
if ( prefix_length > 0 ) { // Just in case it ever changes
2019-11-07 00:03:04 +08:00
ERR_PRINT ( " Prefix for enum ' " _STR ( Error ) " ' is not empty. " ) ;
2017-12-24 10:17:48 +08:00
}
2018-10-19 01:41:10 +08:00
prefix_length = 1 ; // 'ERR_'
2017-12-24 10:17:48 +08:00
}
2018-10-19 01:41:10 +08:00
_apply_prefix_to_enum_constants ( ienum , prefix_length ) ;
2017-12-24 10:17:48 +08:00
}
}
// HARDCODED
List < StringName > hardcoded_enums ;
2020-03-03 17:42:20 +08:00
hardcoded_enums . push_back ( " Vector2.Axis " ) ;
2022-12-07 23:11:39 +08:00
hardcoded_enums . push_back ( " Vector2I.Axis " ) ;
2017-12-24 10:17:48 +08:00
hardcoded_enums . push_back ( " Vector3.Axis " ) ;
2022-12-07 23:11:39 +08:00
hardcoded_enums . push_back ( " Vector3I.Axis " ) ;
2021-07-26 00:45:00 +08:00
for ( const StringName & enum_cname : hardcoded_enums ) {
2017-12-24 10:17:48 +08:00
// These enums are not generated and must be written manually (e.g.: Vector3.Axis)
2018-04-29 04:25:25 +08:00
// Here, we assume core types do not begin with underscore
2017-12-24 10:17:48 +08:00
TypeInterface enum_itype ;
2018-04-29 04:25:25 +08:00
enum_itype . is_enum = true ;
2021-07-26 00:45:00 +08:00
enum_itype . name = enum_cname . operator String ( ) ;
enum_itype . cname = enum_cname ;
2022-12-07 23:11:39 +08:00
enum_itype . proxy_name = pascal_to_pascal_case ( enum_itype . name ) ;
2018-04-29 04:25:25 +08:00
TypeInterface : : postsetup_enum_type ( enum_itype ) ;
2017-12-24 10:17:48 +08:00
enum_types . insert ( enum_itype . cname , enum_itype ) ;
}
}
2019-05-24 06:40:16 +08:00
void BindingsGenerator : : _initialize_blacklisted_methods ( ) {
blacklisted_methods [ " Object " ] . push_back ( " to_string " ) ; // there is already ToString
blacklisted_methods [ " Object " ] . push_back ( " _to_string " ) ; // override ToString instead
blacklisted_methods [ " Object " ] . push_back ( " _init " ) ; // never called in C# (TODO: implement it)
}
2019-04-25 21:45:12 +08:00
void BindingsGenerator : : _log ( const char * p_format , . . . ) {
if ( log_print_enabled ) {
va_list list ;
va_start ( list , p_format ) ;
OS : : get_singleton ( ) - > print ( " %s " , str_format ( p_format , list ) . utf8 ( ) . get_data ( ) ) ;
va_end ( list ) ;
}
}
void BindingsGenerator : : _initialize ( ) {
2019-09-24 21:09:54 +08:00
initialized = false ;
2017-10-03 05:24:00 +08:00
EditorHelp : : generate_doc ( ) ;
2017-12-24 10:17:48 +08:00
enum_types . clear ( ) ;
2019-05-24 06:40:16 +08:00
_initialize_blacklisted_methods ( ) ;
2019-09-24 21:09:54 +08:00
bool obj_type_ok = _populate_object_type_interfaces ( ) ;
ERR_FAIL_COND_MSG ( ! obj_type_ok , " Failed to generate object type interfaces " ) ;
2017-10-03 05:24:00 +08:00
_populate_builtin_type_interfaces ( ) ;
2017-12-24 10:17:48 +08:00
_populate_global_constants ( ) ;
2018-09-04 11:40:41 +08:00
// Generate internal calls (after populating type interfaces and global constants)
2017-12-24 10:17:48 +08:00
2022-05-08 16:09:19 +08:00
for ( const KeyValue < StringName , TypeInterface > & E : obj_types ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
const TypeInterface & itype = E . value ;
Error err = _populate_method_icalls_table ( itype ) ;
ERR_FAIL_COND_MSG ( err ! = OK , " Failed to generate icalls table for type: " + itype . name ) ;
2020-05-16 10:03:05 +08:00
}
2019-09-24 21:09:54 +08:00
initialized = true ;
2017-10-03 05:24:00 +08:00
}
2021-03-13 08:16:27 +08:00
static String generate_all_glue_option = " --generate-mono-glue " ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
static void handle_cmdline_options ( String glue_dir_path ) {
2021-03-13 08:16:27 +08:00
BindingsGenerator bindings_generator ;
bindings_generator . set_log_print_enabled ( true ) ;
if ( ! bindings_generator . is_initialized ( ) ) {
ERR_PRINT ( " Failed to initialize the bindings generator " ) ;
return ;
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
CRASH_COND ( glue_dir_path . is_empty ( ) ) ;
2021-03-13 08:16:27 +08:00
2022-08-30 08:34:01 +08:00
if ( bindings_generator . generate_cs_api ( glue_dir_path . path_join ( API_SOLUTION_NAME ) ) ! = OK ) {
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
ERR_PRINT ( generate_all_glue_option + " : Failed to generate the C# API. " ) ;
2021-03-13 08:16:27 +08:00
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
}
2021-03-13 08:16:27 +08:00
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
static void cleanup_and_exit_godot ( ) {
// Exit once done
Main : : cleanup ( true ) ;
: : exit ( 0 ) ;
2021-03-13 08:16:27 +08:00
}
2017-10-03 05:24:00 +08:00
void BindingsGenerator : : handle_cmdline_args ( const List < String > & p_cmdline_args ) {
2019-07-03 15:44:53 +08:00
String glue_dir_path ;
2021-03-13 08:16:27 +08:00
2017-10-03 05:24:00 +08:00
const List < String > : : Element * elem = p_cmdline_args . front ( ) ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
while ( elem ) {
2019-07-03 15:44:53 +08:00
if ( elem - > get ( ) = = generate_all_glue_option ) {
2017-10-03 05:24:00 +08:00
const List < String > : : Element * path_elem = elem - > next ( ) ;
if ( path_elem ) {
2019-07-03 15:44:53 +08:00
glue_dir_path = path_elem - > get ( ) ;
2017-10-03 05:24:00 +08:00
elem = elem - > next ( ) ;
} else {
2019-11-07 00:03:04 +08:00
ERR_PRINT ( generate_all_glue_option + " : No output directory specified (expected path to '{GODOT_ROOT}/modules/mono/glue'). " ) ;
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
// Exit once done with invalid command line arguments
cleanup_and_exit_godot ( ) ;
2019-07-03 15:44:53 +08:00
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
break ;
2017-10-03 05:24:00 +08:00
}
elem = elem - > next ( ) ;
}
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
if ( glue_dir_path . length ( ) ) {
handle_cmdline_options ( glue_dir_path ) ;
2019-04-25 21:45:12 +08:00
// Exit once done
C#: Move marshaling logic and generated glue to C#
We will be progressively moving most code to C#.
The plan is to only use Mono's embedding APIs to set things at launch.
This will make it much easier to later support CoreCLR too which
doesn't have rich embedding APIs.
Additionally the code in C# is more maintainable and makes it easier
to implement new features, e.g.: runtime codegen which we could use to
avoid using reflection for marshaling everytime a field, property or
method is accessed.
SOME NOTES ON INTEROP
We make the same assumptions as GDNative about the size of the Godot
structures we use. We take it a bit further by also assuming the layout
of fields in some cases, which is riskier but let's us squeeze out some
performance by avoiding unnecessary managed to native calls.
Code that deals with native structs is less safe than before as there's
no RAII and copy constructors in C#. It's like using the GDNative C API
directly. One has to take special care to free values they own.
Perhaps we could use roslyn analyzers to check this, but I don't know
any that uses attributes to determine what's owned or borrowed.
As to why we maily use pointers for native structs instead of ref/out:
- AFAIK (and confirmed with a benchmark) ref/out are pinned
during P/Invoke calls and that has a cost.
- Native struct fields can't be ref/out in the first place.
- A `using` local can't be passed as ref/out, only `in`. Calling a
method or property on an `in` value makes a silent copy, so we want
to avoid `in`.
REGARDING THE BUILD SYSTEM
There's no longer a `mono_glue=yes/no` SCons options. We no longer
need to build with `mono_glue=no`, generate the glue and then build
again with `mono_glue=yes`. We build only once and generate the glue
(which is in C# now).
However, SCons no longer builds the C# projects for us. Instead one
must run `build_assemblies.py`, e.g.:
```sh
%godot_src_root%/modules/mono/build_scripts/build_assemblies.py \
--godot-output-dir=%godot_src_root%/bin \
--godot-target=release_debug`
```
We could turn this into a custom build target, but I don't know how
to do that with SCons (it's possible with Meson).
OTHER NOTES
Most of the moved code doesn't follow the C# naming convention and
still has the word Mono in the names despite no longer dealing with
Mono's embedding APIs. This is just temporary while transitioning,
to make it easier to understand what was moved where.
2021-05-03 21:21:06 +08:00
cleanup_and_exit_godot ( ) ;
2019-04-25 21:45:12 +08:00
}
2017-10-03 05:24:00 +08:00
}
# endif
