In the following commit:
commit 7feb7d068a
Date: Mon May 11 12:08:03 2009 +0000
The last useful uses of normal_mask in linux-nat.c were removed, since
then this variable has sat around being initialised, but never used.
There should be no user visible changes after this commit.
gdb/ChangeLog:
* linux-nat.c (normal_mask): Delete.
(_initialize_linux_nat): Don't initialise normal_mask.
PR 24445 ("dwz multifile index not written to index cache") exposed the
fact that we are not doing things right when we generate an index for an
object file that has is linked to a dwz file. The same happens whether
the index is generated with the intent of populating the index cache or
using the save gdb-index command.
The problem can be observed when running these tests with the
cc-with-dwz-m board:
FAIL: gdb.base/index-cache.exp: test_cache_enabled_hit: check index-cache stats
FAIL: gdb.dwarf2/gdb-index.exp: index used
FAIL: gdb.dwarf2/gdb-index.exp: index used after symbol reloading
When generating the index for such file and inspecting the CU list of the
resulting index (with readelf --debug-dump=gdb_index), we can see something
like:
CU table:
[ 0] 0x0 - 0xb9
[ 1] 0x0 - 0x44
This is supposed to be a sorted list of the ranges of all CUs in the
file this index represents, so already having some overlap is a red
flag. It turns out that we save the ranges of CUs coming from both the
main file and the dwz file in the same index.
After digging a little bit, it became quite obvious that the index in
the main file should only list the CUs present in the main file, and a
separate index should be generated for the dwz file, listing the CUs
present in that file.
First, that's what happens if you run dwz on a file that already has a
GDB index embedded. Second, dwarf2read.c has code to read an index from
a dwz file. The index in the dwz file is actually required to be
present, if the main file has an index.
So this patch changes write_psymtabs_to_index to generate an index for
the dwz file, if present. That index only contains a CU list, just like
what the dwz tool does when processing a file that already contains an
index.
Some notes about the implementation:
- The file management (creating a temp file, make sure it's
close/removed on error - in the right order) is a bit heavy in
write_psymtabs_to_index, and I needed to add a third file. I factored
this pattern in a separate class, index_wip_file.
- It became a bit tedious to keep the call to assert_file_size in
write_psymtabs_to_index, write_gdbindex would have had to return two
sizes. Instead, I moved the calls to assert_file_size where the file
is written. The downside is that we lose the filename at this point,
but it was only used for the very improbable case of ftell failing, so
I think it's not a problem.
- The actual writing of the index file is factored out to
write_gdbindex_1, so it can be re-used for both index files.
- While the "save gdb-index" command will now write two .gdb-index
files, this patch does not update the gdb-add-index.sh script, this
will come in a later patch.
gdb/ChangeLog:
YYYY-MM-DD Simon Marchi <simon.marchi@efficios.com>
PR gdb/24445
* dwarf-index-write.h (write_psymtabs_to_index): Add
dwz_basename parameter.
* dwarf-index-write.c (write_gdbindex): Move file writing to
write_gdbindex_1. Change return type void.
(assert_file_size): Move up, remove filename parameter.
(write_gdbindex_1): New function.
(write_debug_names): Change return type to void, call
assert_file_size.
(struct index_wip_file): New struct.
(write_psymtabs_to_index): Add dwz_basename parameter. Move
file logic to index_wip_file. Write index for dwz file if
needed.
(save_gdb_index_command): Pass basename of dwz file, if present.
* dwarf-index-cache.c (index_cache::store): Obtain and pass
build-id of dwz file, if present.
* dwarf2read.c (struct dwz_file): Move to dwarf2read.h.
(dwarf2_get_dwz_file): Likewise.
* dwarf2read.h (struct dwz_file): Move from dwarf2read.c.
(dwarf2_get_dwz_file): Likewise.
gdb/testsuite/ChangeLog:
YYYY-MM-DD Tom de Vries <tdevries@suse.de>
PR gdb/24445
* gdb.dwarf2/gdb-index.exp (add_gdb_index): Update dwz file with
generated index.
I noticed a couple of spots using concat that could use xstrdup
instead. This patch fixes these.
gdb/ChangeLog
2019-06-16 Tom Tromey <tom@tromey.com>
* coffread.c (process_coff_symbol): Use xstrdup.
* value.c (create_internalvar): Use xstrdup.
In bug gdb/24686 a testsuite failure was reported, this failure was
actually just another instance of bug gdb/24541, however, due to the
non-deterministic nature of bug gdb/24541 the testsuite pattern that
was intended to catch this bug failed.
This commit adds a second pattern to help detect gdb/24541, which
should change the FAIL reported in gdb/24686 into a KFAIL.
gdb/testsuite/ChangeLog:
PR gdb/24686
* gdb.mi/mi-catch-cpp-exceptions.exp: Add an extra pattern to
improve detection of bug gdb/24541.
I found a few spots in the TUI that were NULL-checking the result of
XNEW. This cannot return NULL, so this patch removes the checks.
gdb/ChangeLog
2019-06-16 Tom Tromey <tom@tromey.com>
* tui/tui-data.c (tui_alloc_generic_win_info)
(tui_alloc_win_info, tui_add_content_elements): Remove NULL
checks.
The gdb.base/index-cache.exp test fails with the readnow board:
$ make check TESTS="gdb.base/index-cache.exp" RUNTESTFLAGS="--target_board=readnow"
FAIL: gdb.base/index-cache.exp: test_cache_enabled_miss: at least one file was created
FAIL: gdb.base/index-cache.exp: test_cache_enabled_miss: expected file is there
FAIL: gdb.base/index-cache.exp: test_cache_enabled_miss: check index-cache stats
FAIL: gdb.base/index-cache.exp: test_cache_enabled_hit: check index-cache stats
The problem is similar to what was fixed in
5a56d6a65f ("[gdb/testsuite] Fix index-cache.exp with cc-with-{gdb-index,debug-names}")
In that commit, gdb.base/index-cache.exp was modified to account for the
fact that the index cache is not used when the binary already has an
embedded index.
The same situation happens when GDB is started with the -readnow flag:
it bypasses indices and partial symbols. So this patch updates the test
to also expect the index cache not to be used if -readnow is present in
$GDBFLAGS,
gdb/testsuite/ChangeLog:
PR gdb/24669
* gdb.base/index-cache.exp (uses_readnow,
expecting_index_cache_use): Define global variable.
(test_cache_enabled_miss, test_cache_enabled_hit): Use
expecting_index_cache_use.
Show the type of not-allocated and/or not-associated types. For array
types and pointer to array types we are going to print the number of
ranks.
Consider this Fortran program:
program test
integer, allocatable :: vla (:)
logical l
allocate (vla(5:12))
l = allocated (vla)
end program test
And this GDB session with current HEAD:
(gdb) start
...
2 integer, allocatable :: vla (:)
(gdb) n
4 allocate (vla(5:12))
(gdb) ptype vla
type = <not allocated>
(gdb) p vla
$1 = <not allocated>
(gdb)
And the same session with this patch applied:
(gdb) start
...
2 integer, allocatable :: vla (:)
(gdb) n
4 allocate (vla(5:12))
(gdb) ptype vla
type = integer(kind=4), allocatable (:)
(gdb) p vla
$1 = <not allocated>
(gdb)
The type of 'vla' is now printed correctly, while the value itself
still shows as '<not allocated>'. How GDB prints the type of
associated pointers has changed in a similar way.
gdb/ChangeLog:
* f-typeprint.c (f_print_type): Don't return early for not
associated or not allocated types.
(f_type_print_varspec_suffix): Add print_rank parameter and print
ranks of array types in case they dangling.
(f_type_print_base): Add print_rank parameter.
gdb/testsuite/ChangeLog:
* gdb.fortran/pointers.f90: New file.
* gdb.fortran/print_type.exp: New file.
* gdb.fortran/vla-ptype.exp: Adapt expected results.
* gdb.fortran/vla-type.exp: Likewise.
* gdb.fortran/vla-value.exp: Likewise.
* gdb.mi/mi-vla-fortran.exp: Likewise.
Adds some MI commands to catch C++ exceptions. The new commands are
-catch-throw, -catch-rethrow, and -catch-catch, these all correspond
to the CLI commands 'catch throw', 'catch rethrow', and 'catch catch'.
Each MI command takes two optional arguments, '-t' has the effect of
calling 'tcatch' instead of 'catch', for example:
(gdb)
-catch-throw -t
Is the same as:
(gdb) tcatch throw
There is also a '-r REGEXP' argument that can supply a regexp to match
against the exception type, so:
(gdb)
-catch-catch -r PATTERN
Is the same as:
(gdb) catch catch PATTERN
The change in print_mention_exception_catchpoint might seem a little
strange; changing the output from using ui_out::field_int and
ui_out::text to using ui_out::message.
The print_mention_exception_catchpoint is used as the 'print_mention'
method for the exception catchpoint breakpoint object. Most of the
other 'print_mention' methods (see breakpoint.c) use either
printf_filtered, of ui_out::message. Using field_int was causing an
unexpected field to be added to the MI output. Here's the output
without the change in print_mention_exception_catchpoint:
(gdb)
-catch-throw
^done,bkptno="1",bkpt={number="1",type="breakpoint",disp="keep",
enabled="y",addr="0x00000000004006c0",
what="exception throw",catch-type="throw",
thread-groups=["i1"],times="0"}
Notice the breakpoint number appears in both the 'bkptno' field, and
the 'number' field within the 'bkpt' tuple. Here's the output with
the change in print_mention_exception_catchpoint:
(gdb)
-catch-throw
^done,bkpt={number="1",type="breakpoint",disp="keep",
enabled="y",addr="0x00000000004006c0",
what="exception throw",catch-type="throw",
thread-groups=["i1"],times="0"}
gdb/ChangeLog:
* NEWS: Mention new MI commands.
* break-catch-throw.c (enum exception_event_kind): Move to
breakpoint.h.
(print_mention_exception_catchpoint): Output text as a single
message.
(catch_exception_command_1): Rename to...
(catch_exception_event): ...this, make non-static, update header
command, and change some parameter types.
(catch_catch_command): Update for changes to
catch_exception_command_1.
(catch_throw_command): Likewise.
(catch_rethrow_command): Likewise.
* breakpoint.c (enum exception_event_kind): Delete.
* breakpoint.h (enum exception_event_kind): Moved here from
break-catch-throw.c.
(catch_exception_event): Declare.
* mi/mi-cmd-catch.c (mi_cmd_catch_exception_event): New function.
(mi_cmd_catch_throw): New function.
(mi_cmd_catch_rethrow): New function.
(mi_cmd_catch_catch): New function.
* mi/mi-cmds.c (mi_cmds): Add 'catch-throw', 'catch-rethrow', and
'catch-catch' entries.
* mi/mi-cmds.h (mi_cmd_catch_throw): Declare.
(mi_cmd_catch_rethrow): Declare.
(mi_cmd_catch_catch): Declare.
gdb/doc/ChangeLog:
* gdb.texinfo (GDB/MI Catchpoint Commands): Add menu entry to new
node.
(C++ Exception GDB/MI Catchpoint Commands): New node to describe
new MI commands.
gdb/testsuite/ChangeLog:
* gdb.mi/mi-catch-cpp-exceptions.cc: New file.
* gdb.mi/mi-catch-cpp-exceptions.exp: New file.
* lib/mi-support.exp (mi_expect_stop): Handle 'exception-caught'
as a stop reason.
The annotations should be additional information printed by GDB to be
consumed by users (GUIs), but GDB shouldn't reduce what it prints
based on whether annotations are on or not. However, this is what
happens for annotate_source_line.
This commit makes annotate_source_line a void function that simply
outputs the annotation information, GDB will then print the contents
of the source line to the terminal in the normal way.
Some tests needed to be updated after this commit.
gdb/ChangeLog:
* annotate.c (annotate_source_line): Change return type to void,
update implementation to match.
* annotate.h (annotate_source_line): Change return type to void,
update header comment.
* stack.c (print_frame_info): Don't change what frame information
is printed based on whether annotations are on or not.
gdb/testsuite/ChangeLog:
* gdb.base/annota1.exp: Update expected results.
* gdb.cp/annota2.exp: Likewise.
* gdb.cp/annota3.exp: Likewise.
While reviewing some of the annotation code I noticed that
identify_source_line (in source.c) sets current_source_line,
current_source_symtab, and also calls clear_lines_listed_range. This
seems a little strange, identify_source_line is really a wrapper
around annotate_source, and is only called when annotation_level is
greater than 0 (so annotations are turned on).
It seems weird (to me) that when annotations are on we update GDB's
idea of the "current" line/symtab, but when they are off we don't,
given that annotations are really about communicating GDB's state to a
user (GUI) and surely shouldn't be changing GDB's behaviour.
This commit removes from identify_source_line all of the setting of
current line/symtab and the call to clear_lines_listed_range, after
doing this GDB still passes all tests, so I don't believe these lines
were actually required.
With this code removed identify_source_line is only a wrapper around
annotate_source, so I moved identify_source_line to annotate.c and
renamed it to annotate_source_line.
gdb/ChangeLog:
* annotate.c: Add 'source.h' and 'objfiles.h' includes.
(annotate_source): Make static.
(annotate_source_line): Moved from source.c and renamed from
identify_source_line. Update the return type.
* annotate.h (annotate_source): Delete declaration.
(annotate_source_line): Declaration moved from source.h, and
renamed from identify_source_line. Return type updated.
* source.c (identify_source_line): Moved to annotate.c and renamed
to annotate_source_line.
(info_line_command): Remove check of annotation_level.
* source.h (identify_source_line): Move declaration to annotate.h
and rename to annotate_source_line.
* stack.c: Add 'annotate.h' include.
(print_frame_info): Remove check of annotation_level before
calling annotate_source_line.
Every place that a symtab's line_charpos data is loaded always follows
the same pattern, so create a new function to contain this logic and
make use of it throughout GDB.
There should be no user visible changes after this commit.
gdb/ChangeLog:
* source-cache.c (source_cache::get_plain_source_lines): Use
open_source_file_with_line_charpos instead of just
open_source_file, remove call to find_source_lines.
(source_cache::get_source_lines): Likewise.
* source.c (find_source_lines): Make static.
(get_filename_and_charpos): Renamed into...
(open_source_file_with_line_charpos): ..this along with changes to
return a scoped_fd, and some other minor clean ups.
(identify_source_line): Use open_source_file_with_line_charpos.
(search_command_helper): Use open_source_file_with_line_charpos
instead of just open_source_file, remove call to
find_source_lines.
* source.h (open_source_file_with_line_charpos): Declare new
function.
(find_source_lines): Delete declaration.
The parameter 'fullname' is always passed as NULL to the function
get_filename_and_charpos in source.c, so lets remove the parameter.
There should be no user visible changes after this commit.
gdb/ChangeLog:
* source.c (get_filename_and_charpos): Remove fullname
parameter.
(identify_source_line): Update call to get_filename_and_charpos.
Having paths in test names makes comparing sum files difficult, rename
a test to avoid paths in test names.
gdb/testsuite/ChangeLog:
* gdb.base/style-logging.exp: Remove path from test name.
PR gdb/24502 requests that the "set logging" log file not contain
style escape sequences emitted by gdb.
This seemed like a reasonable request to me, so this patch implements
filtering for the log file.
This also updates a comment in ui-style.h that I noticed while writing
the patch.
Tested on x86-64 Fedora 29.
gdb/ChangeLog
2019-06-14 Tom Tromey <tromey@adacore.com>
PR gdb/24502:
* ui-style.h (skip_ansi_escape): Update comment.
* ui-file.h (class no_terminal_escape_file): New class.
* ui-file.c (no_terminal_escape_file::write)
(no_terminal_escape_file::puts): New methods.
* cli/cli-logging.c (handle_redirections): Use
no_terminal_escape_file.
gdb/testsuite/ChangeLog
2019-06-14 Tom Tromey <tromey@adacore.com>
PR gdb/24502:
* gdb.base/style-logging.exp: New file.
I noticed that a NEWS item about Python scripting changes appeared
between an item about a convenience function and an item about a
convenience variable. I think it's better for the latter to be next
to each other.
gdb/ChangeLog
2019-06-14 Tom Tromey <tromey@adacore.com>
* NEWS: Move convenience variable news above Python news.
This patch moves the gdb/gnulib subdirectory to the top level.
It adjusts the top-level build system to build gnulib when necessary,
and changes gdb to use this. However, gdbserver still builds its own
copy of gnulib, just from the new source location.
A small hack was needed to ensure that gnulib is only built when gdb
is enabled. The Makefile only provides an ordering -- the directory
must be mentioned in configdirs to actually be compiled at all.
Most of the patch is just a "git mv" of gnulib, though a few minor
path adjustments were needed in some files there.
Tested by the buildbot.
ChangeLog
2019-06-14 Tom Tromey <tom@tromey.com>
* MAINTAINERS: Add gnulib.
* gnulib: New directory, move from gdb/gnulib.
* configure.ac (host_libs): Add gnulib.
* configure: Rebuild.
* Makefile.def (host_modules, dependencies): Add gnulib.
* Makefile.in: Rebuild.
gdb/ChangeLog
2019-06-14 Tom Tromey <tom@tromey.com>
* gnulib: Move directory to top-level.
* configure.ac: Don't configure gnulib.
* configure: Rebuild.
* common/common-defs.h: Use new path to gnulib.
* Makefile.in (GNULIB_BUILDDIR): Now ../gnulib.
(GNULIB_H): Remove.
(INCGNU): Look in new gnulib location.
(HFILES_NO_SRCDIR): Remove gnulib files.
(SUBDIR, REQUIRED_SUBDIRS): Remove gnulib.
(generated_files): Remove GNULIB_H.
($(LIBGNU), all-lib): Remove targets.
(distclean): Don't mention GNULIB_BUILDDIR.
($(GNULIB_BUILDDIR)/Makefile): Remove target.
gdb/gdbserver/ChangeLog
2019-06-14 Tom Tromey <tom@tromey.com>
* configure.ac: Use new path to gnulib.
* configure: Rebuild.
* Makefile.in (INCGNU, $(GNULIB_BUILDDIR)/Makefile): Use new path
to gnulib.
gnulib/ChangeLog
2019-06-14 Tom Tromey <tom@tromey.com>
* update-gnulib.sh: Adjust paths.
* Makefile.in: Adjust paths.
* configure.ac: Adjust paths. Use ACX_LARGEFILE.
* configure: Rebuild.
A user suggested that add-symbol-file ought to warn if the file does
not in fact provide any symbols. This seemed like a decent idea, so
this patch implements this idea.
Tested on x86-64 Fedora 29.
gdb/ChangeLog
2019-06-14 Tom Tromey <tromey@adacore.com>
* symfile.c (add_symbol_file_command): Remove obsolete comment.
Warn if symbol file does not provide any symbols.
gdb/testsuite/ChangeLog
2019-06-14 Tom Tromey <tromey@adacore.com>
* gdb.base/symfile-warn.exp: New file.
* gdb.base/symfile-warn.c: New file.
A user noted that, when sources are symlinked, gdb annotations will
print the real path, rather than the name of the symlink.
It seems to me that it is better to print the name of the file that
was actually used in the build, unless there is some reason not to.
This patch implements this, with the caveat that it will not work when
basenames-may-differ is enabled. The way this mode is currently
implemented, returning the symbolic (not real) path is not possible.
While I think it would be good to redo the source file name cache and
perhaps integrate it with class source_cache, I haven't done so here.
Regression tested on x86-64 Fedora 29.
gdb/ChangeLog
2019-06-14 Tom Tromey <tromey@adacore.com>
* source.c (find_and_open_source): Respect basenames_may_differ.
gdb/testsuite/ChangeLog
2019-06-14 Tom Tromey <tromey@adacore.com>
* gdb.base/annotate-symlink.exp: New file.
In a couple of places in annotate.c we are manually backing up and
restoring the terminal ownership, we could instead make use of
scoped_restore_terminal_state.
gdb/ChangeLog:
* annotate.c (annotate_breakpoints_invalid): Make use of
scoped_restore_terminal_state.
(annotate_frames_invalid): Likewise.
In Ada mode, re-assigning an array of a different size to a
convenience variable will cause an error:
(gdb) set lang ada
(gdb) set $v := "abc"
(gdb) set $v := "abcd"
cannot assign arrays of different length
However, this does not really make sense -- instead, it should always
be possible to overwrite a convenience variable.
This patch fixes this bug.
This was reviewed off-list by Joel. I'm checking it in.
gdb/ChangeLog
2019-06-14 Tom Tromey <tromey@adacore.com>
* ada-lang.c (ada_evaluate_subexp) <case BINOP_ASSIGN>: Always
allow assignment to an internalvar.
gdb/testsuite/ChangeLog
2019-06-14 Tom Tromey <tromey@adacore.com>
* gdb.ada/set_wstr.exp: Add reassignment test.
Ada attribute names can contain "_", but the lexer currently does not
allow this -- even though the "attributes" array lists some attributes
spelled this way.
This patch fixes the bug and adds test cases for the existing
attributes.
This was reviewed off-list by Joel. I'm checking it in.
gdb/ChangeLog
2019-06-14 Tom Tromey <tromey@adacore.com>
* ada-lex.l: Allow "_" in attribute names.
gdb/testsuite/ChangeLog
2019-06-14 Tom Tromey <tromey@adacore.com>
* gdb.ada/formatted_ref.exp (test_p_x_addr): Check
'unchecked_access and 'unrestricted_access as well.
PR gdb/24653 points out that a gcc snapshot will complain about the
calls to alloca(0) in gdb.
These calls are only needed when using the C alloca. I'm inclined to
think that no current platform needs this, so this patch removes the
calls.
Let me know what you think.
gdb/ChangeLog
2019-06-14 Tom Tromey <tromey@adacore.com>
PR gdb/24653:
* regcache.c (registers_changed): Don't call alloca.
* top.c (execute_command): Don't call alloca.
PR ada/24539 concerns a test failure in gdb.ada/vla.exp.
The problem here is that different versions of Gnat emit the
structure's fields in different orders -- with the order currently
failing actually being the correct one.
Joel pointed out that this can be fixed by simply adding the
No_Component_Reordering pragma to the type in question, which is what
this patch does.
I've reported a Gnat compiler bug internally in hopes of getting the
underlying problem fixed.
gdb/testsuite/ChangeLog
2019-06-14 Tom Tromey <tromey@adacore.com>
PR ada/24539:
* gdb.ada/vla.exp: Update tests.
* gdb.ada/vla/vla.adb (Record_Type): Use No_Component_Reordering
pragma.
These ilp32 relocations were missing for some reason.
bfd/ChangeLog:
* elfnn-aarch64.c: Enable MOVW_PREL relocs for ELF32.
include/ChangeLog:
* elf/aarch64.h (R_AARCH64_P32_MOVW_PREL_G0): Define.
(R_AARCH64_P32_MOVW_PREL_G0_NC): Define.
(R_AARCH64_P32_MOVW_PREL_G1): Define.
ld/ChangeLog:
* testsuite/ld-aarch64/aarch64-elf.exp: Add emit-relocs-22 and -23.
* testsuite/ld-aarch64/emit-relocs-22.d: New test.
* testsuite/ld-aarch64/emit-relocs-22.s: New test.
* testsuite/ld-aarch64/emit-relocs-23.d: New test.
* testsuite/ld-aarch64/emit-relocs-23.s: New test.
"paddi rt,sym@pcrel" as an abbreviation for "paddi rt,0,sym@pcrel,1"
is invalid, so replace with "pla rt,sym@pcrel" which is a valid form
of "pla rt,sym@pcrel(0),1".
* elf64-ppc.c: Fix comments involving paddi.
With integer commands that support "unlimited", we currently fail to
notice junk after "unlimited":
(gdb) show print elements
Limit on string chars or array elements to print is 200.
(gdb) set print elements unlimited foo
(gdb) show print elements
Limit on string chars or array elements to print is unlimited.
This commit fixes that. After, we get:
(gdb) set print elements unlimited foo
Junk after "unlimited": foo
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* cli/cli-setshow.c (cli/cli-setshow.c): New parameter
'expression'. When parsing an expression, error out if there's
junk after "unlimited".
(parse_cli_var_uinteger, parse_cli_var_zuinteger_unlimited)
(do_set_command): Adjust calls to is_unlimited_literal.
gdb/testsuite/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* gdb.base/settings.exp (test-integer): Test junk after
"unlimited".
This is a simple test to ensure that the l.adrp instruction can be assembled and
simulated correctly.
sim/testsuite/sim/or1k/ChangeLog:
yyyy-mm-dd Stafford Horne <shorne@gmail.com>
* adrp.S: New file.
This is a very basic test but it ensure the machine is wired up
correctly and that the assembler works.
sim/testsuite/sim/or1k/ChangeLog:
yyyy-mm-dd Stafford Horne <shorne@gmail.com>
* fpu64a32.S: New file.
Define and wire up unordered floating point comparison operations for cgen
targets. This patch depends on my posted cgen patches[0].
[0] https://www.sourceware.org/ml/cgen/2019-q2/msg00013.html
sim/common/ChangeLog:
yyyy-mm-dd Stafford Horne <shorne@gmail.com>
* cgen-accfp.c (unorderedsf, unordereddf): New functions.
(cgen_init_accurate_fpu): Wire up unorderedsf and unordereddf.
* cgen-fpu.h (cgen_fp_ops): Define fields unorderedsf and unordereddf.
Up until now these have not been used in any CGEN targets, add them as
they are now used by OpenRISC.
sim/common/ChangeLog:
* cgen-accfp.c (floatdidf, fixdfdi): New functions.
(cgen_init_accurate_fpu): Add floatdidf and fixdfdi.
The options framework series broken the build with gcc 4.8, or any
other compiler were we end up forcing -std=gnu++11, causing errors
like these:
../../binutils-gdb/gdb/compile/compile.c: In function gdb::option::option_def_group make_compile_options_def_group(compile_options*):
../../binutils-gdb/gdb/compile/compile.c:266:44: error: could not convert (const gdb::option::option_def*)(& compile_command_option_defs) from const gdb::option::option_def* to gdb::array_view<const gdb::option::option_def>
return {compile_command_option_defs, opts};
^
CXX copying.o
../../binutils-gdb/gdb/compile/compile.c:267:1: error: control reaches end of non-void function [-Werror=return-type]
}
^
This is a C++11 vs C++14 difference -- C++14 relaxed the rules for
eliding braces.
This commit fixes it by adding the missing (in C++11) braces. Tested
with g++ 4.8.
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* compile/compile.c (make_compile_options_def_group): Add braces
around array_view initializer.
* thread.c (make_thread_apply_all_options_def_group)
(make_thread_apply_all_options_def_group): Likewise.
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* NEWS (New commands): Mention "maint test-options
require-delimiter", "maint test-options unknown-is-error", "maint
test-options unknown-is-operand" and "maint show
test-options-completion-result".
(New command options, command completion): New section.
(Completion improvements): New section.
Mention that you can abbreviate "unlimited".
gdb/doc/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* gdb.texinfo (Command Completion): Mention command options too.
(Command Options): New node.
(Threads): Add anchors. Extend descriptions of the "taas" and
"tfaas" commands.
(Backtrace): Describe new options of the "backtrace" command. Add
anchors.
(Frame Apply): Describe new options of the "frame apply" and
"faas" commands. Add anchors.
(Data): Describe new options of the "print" command. Add anchors.
(Compiling and Injecting Code): Mention options of the "compile
print" command.
(Maintenance Commands): Mention "maint test-options" subcommands
and the "maint show test-options-completion-result" command.
Now that "thread/frame apply" have been converted to the gdb::option
framework, these functions are no longer used.
For a while, I thought about keeping the unit tests, by making a local
version of parse_flags_qcs in the unit tests file. But all that would
really test that is used by GDB itself, is the validate_flags_qcs
function. So in the end, I went through all the unit tests, and
converted any that wasn't already covered to gdb.base/options.exp
tests. And those have all already been added in previous patches.
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* cli/cli-utils.c (parse_flags, parse_flags_qcs): Delete.
* cli/cli-utils.h (parse_flags, parse_flags_qcs): Delete.
* unittests/cli-utils-selftests.c (test_parse_flags)
(test_parse_flags_qcs): Delete.
(test_cli_utils): Don't call deleted functions.
Similarly to the "frame apply" patch, this makes the "thread apply"
family of commands -- "thread apply TID", "thread apply all" and
"taas" use the gdb::option framework for '-'-style options.
No new options are added, but there are some user-visible changes:
- Can now abbreviate and complete "-ascending"
- We now have a completer for "thread apply" commands
Can now complete options ("thread apply all -[TAB]"), and also,
'thread apply all COMMAND[TAB]' now does what you'd expect, by
making use of the new complete_command routine.
- "help" output tweaked with auto-generated option descriptions:
~~~
Usage: thread apply all [OPTION]... COMMAND
Prints per-inferior thread number and target system's thread id
followed by COMMAND output.
By default, an error raised during the execution of COMMAND
aborts "thread apply".
Options:
-ascending
Call COMMAND for all threads in ascending order.
The default is descending order.
-q
Disables printing the thread information.
-c
Print any error raised by COMMAND and continue.
-s
Silently ignore any errors or empty output produced by COMMAND.
~~~
The "By default ..." sentence is new as well.
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* thread.c: Include "cli/cli-option.h".
(tp_array_compar_ascending): Global.
(tp_array_compar): Delete function.
(tp_array_compar_ascending, tp_array_compar_descending): New
functions.
(ascending_option_def, qcs_flag_option_def)
(thr_qcs_flags_option_defs)
(make_thread_apply_all_options_def_group)
(make_thread_apply_options_def_group): New.
(thread_apply_all_command): Use gdb::option::process_options.
(thread_apply_command_completer)
(thread_apply_all_command_completer): New.
(thread_apply_command): Use gdb::option::process_options.
(_initialize_thread): Delete THREAD_APPLY_FLAGS_HELP, replace it
with a new THREAD_APPLY_OPTION_HELP. Use gdb::option::build_help
to generate help text of "thread apply". Adjust "taas"'s help.
* tid-parse.c (tid_range_parser::in_thread_range): New method.
* tid-parse.h (tid_range_parser::in_thread_range): New method.
gdb/testsuite/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* gdb.base/options.exp (test-thread-apply): New.
(top level): Call it.
With the following patch, we'll be able to explicitly tell "thread
apply" where options end, using the "--" delimiter. A test added by
that patch caught a pre-existing inconsistency:
(gdb) thread apply 1 -- -
Invalid thread ID: -
(gdb) frame apply level 0 -- -
#0 main () at threads.c:55
Cannot enable the TUI when output is not a terminal
Above, "thread apply" did not try to run the command, while "frame
apply level" did. ("-" is a valid TUI command.)
That "-" is past "--", so it should have not been confused with an
invalid TID, in the "thread apply" case.
That error actually doesn't come from the TID parser, but instead from
thread_apply_command directly.
So that error/check needs tweaking. The next question is what to
tweak it to.
"-" is actually a valid TUI command:
(gdb) help -
Scroll window backward.
Usage: - [WIN] [N]
(gdb) frame apply level 0 -- -
#0 main () at threads.c:55
Cannot enable the TUI when output is not a terminal
While I don't imagine it being useful to use that "-" command with
"thread apply" or "frame apply level", the fact is that you can use it
with "frame apply level", but not with "thread apply". And since it's
an actual command, pedantically it seems right to allow it.
That's what this commit does.
Note: simply removing the "isalpha" check regresses
gdb.multi/tids.exp -- see related commit 3f5b759880.
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* thread.c (thread_apply_command): Check for invalid TID with
isdigit instead of !isalpha.
This adds support for '-'-style options to the "frame apply" family of
commands -- "frame apply COUNT", "frame apply level", "frame apply
all", "faas" and "tfaas".
The -q/-c/-s flags were already supported, -past-main/-past-entry is
new:
~~~
(gdb) help frame apply all
Apply a command to all frames.
Usage: frame apply all [OPTION]... COMMAND
Prints the frame location information followed by COMMAND output.
By default, an error raised during the execution of COMMAND
aborts "frame apply".
Options:
-q
Disables printing the frame location information.
-c
Print any error raised by COMMAND and continue.
-s
Silently ignore any errors or empty output produced by COMMAND.
-past-main [on|off]
Set whether backtraces should continue past "main".
Normally the caller of "main" is not of interest, so GDB will terminate
the backtrace at "main". Set this if you need to see the rest
of the stack trace.
-past-entry [on|off]
Set whether backtraces should continue past the entry point of a program.
Normally there are no callers beyond the entry point of a program, so GDB
will terminate the backtrace there. Set this if you need to see
the rest of the stack trace.
~~~
TAB completion of options is now supported. Also, TAB completion of
COMMAND in "frame apply all COMMAND" does the right thing now, making
use of complete_command, added by the previous patch. E.g.:
(gdb) thread apply all -ascending frame apply all -past-main print -[TAB]
-address -elements -pretty -symbol
-array -null-stop -repeats -union
-array-indexes -object -static-members -vtbl
(gdb) thread apply all -ascending frame apply all -past-main print glo[TAB]
global1 global2
The change to tfaas_command is necessary because otherwise you get
this:
(gdb) tfaas --
Unrecognized option at: frame apply all -s --
That's because the above is equivalent to:
(gdb) thread apply all -s frame apply all -s --
and the "--" instructs "thread apply" to consider everything up to
"--" as its command options. And from that view, "frame" is an
invalid option.
The change makes tfaas be equivalent to:
(gdb) thread apply all -s -- frame apply all -s --
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* cli/cli-utils.c (parse_flags_qcs): Use validate_flags_qcs.
(validate_flags_qcs): New.
* cli/cli-utils.h (struct qcs_flags): Change field types to int.
(validate_flags_qcs): Declare.
* stack.c (qcs_flag_option_def, fr_qcs_flags_option_defs): New.
(make_frame_apply_options_def_group): New.
(frame_apply_command_count): Process options with
gdb::option::process_options.
(frame_apply_completer): New.
(frame_apply_level_completer, frame_apply_all_completer)
(frame_apply_completer): New.
(_initialize_stack): Update help of "frame apply", "frame apply
level", "frame apply all" and "faas" to mention supported options
and install command completers.
* stack.h (frame_apply_all_completer): Declare.
* thread.c: Include "stack.h".
(tfaas_command): Add "--".
(_initialize_thread): Update help "tfaas" to mention supported
options and install command completer.
gdb/testsuite/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* gdb.base/options.exp (test-frame-apply): New.
(top level): Test print commands with different "frame apply"
prefixes.
This adds a completion helper routine that makes it possible for a
command that takes another command as argument, such as "frame apply
all COMMAND" as "thread apply all COMMAND", to complete on COMMAND,
and have the completion machinery recurse and complete COMMAND as if
you tried to complete "(gdb) COMMAND". I.e., we'll be able to
complete like this, for example:
(gdb) thread apply all -[TAB]
-c -ascending -q -s
(gdb) thread apply all -ascending frame apply all -[TAB]
-c -limit -past-entry -past-main -q -s
(gdb) thread apply all -ascending frame apply all -past-main print -[TAB]
-address -elements -pretty -symbol
-array -null-stop -repeats -union
-array-indexes -object -static-members -vtbl
(gdb) thread apply all -ascending frame apply all -past-main print glo[TAB]
global1 global2
Above, the completer function understands that "thread apply all" is a
command, and then parses "-ascending" successfully and understand that
the rest of the string is "thread apply all"'s operand. And then, the
process repeats for the "frame apply" command, and on and on.
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* completer.c (complete_nested_command_line): New.
(gdb_completion_word_break_characters_throw): Add assertion.
* completer.h (complete_nested_command_line): Declare.
This adds a procedure to the collection of completion-testing
routines, that allows checking whether completion offers all commands
as completion candidates. This will be used for testing completing
"frame apply all [TAB]", "thread apply all [TAB]", etc.
gdb/testsuite/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* lib/completion-support.exp (test_gdb_complete_tab_multiple)
(test_gdb_complete_cmd_multiple, test_gdb_complete_multiple): Add
'max_completions' parameter and handle it.
(test_gdb_completion_offers_commands): New.
"backtrace"'s completer now completes on command options:
(gdb) bt -[TAB]
-entry-values -full -no-filters -past-main
-frame-arguments -hide -past-entry -raw-frame-arguments
But it doesn't know how to complete on qualifiers:
(gdb) bt fu[TAB]
funlockfile futimens futimes.c
funlockfile.c futimens.c futimesat
futex-internal.h futimes futimesat.c
This commit fixes that:
(gdb) bt fu[TAB]ll
(gdb) bt n[TAB]o-filters
(gdb) bt h[TAB]ide
I considered teaching the gdb::option framework to handle non-'-'
options, but decided it wasn't worth it for this special case, and I'd
rather not make it easy to add new qualifier-like options.
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* stack.c (parse_backtrace_qualifiers): New.
(backtrace_command): Use it.
(backtrace_command_completer): Complete on qualifiers.
gdb/testsuite/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* gdb.base/options.exp (test-backtrace): Test completing qualifiers.
