malloc(0) can legitimately return NULL; it does on some systems and
not others. Force the size to 1 byte if the size is 0 coming in,
except for realloc() where this is legitimate and equivalent to
free().
Since this is an abnormal case, and can't even happen with most C
libraries, handle it on the error path, after we already got back a
NULL pointer.
Reported-by: Ozkan Sezer <sezeroz@gmail.com>
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Clean up some perl warnings, some of which were legitimate (apparently
undef doesn't actually take a list of arguments, a common enough
mistake that it is mentioned in the man page!, and a list of variables
after "my" can be cantankerous), and some of which were nuisance but
were easy enough to clean up.
Maybe this can resolve the problems with very old version of Perl?
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
The single-line macro argument parsing was completely broken as a
comma would not be recognized as an argument separator.
In the process of fixing this, make a fair bit of code cleanups.
Note: reverse tokens for smacro->expansion doesn't actually make any
sense anymore, might reconsider that.
This checkin also removes the distinction between "magic" and plain
smacros; the only difference is which specific expand method is being
invoked.
Finally, extend the allocating-string functions such that *all* the
allocating string functions support querying the length of the string
a posteori.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Windows supports pathnames up to 32767 UTF-16 characters, but using
the standard interfaces only up to 260 characters. Wrap the functions
that take filenames on Windows.
Clean up the compatiblity layers some more for reduced #ifdefs.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
The smacro expansion code was virtually impossible to understand, and
was leading to very strange failures. Clean it up, and do much better
handling of magic macros. This should also allow for recursive
macros, but recursive macros are extremely tricky in that it is very
hard to keep them from recursing forever, unless there is at least one
argument which is never expanded. They are not currently implemented.
Even so, I believe token pasting makes it possible to create infinite
loops; e.g.:
%define foo foo %+
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
"compiler.h" already includes a bunch of common include files. There
is absolutely no reason to duplicate them in individual files, and in
fact it robs us of central control of how these files are used.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
For almost everything we should use "nctype.h". Right now we don't
have a nasm_toupper() to use <ctype.h> for things that need toupper().
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
There is absolutely no reason not to include <string.h> globally, and
with the inline function for mempcpy() we need it there anyway.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
With buffered warnings, most warnings *must* be issued on every pass,
so ERR_PASS1 is simply wrong in most cases.
ERR_PASS1 now means "force this warning to be output even in
pass_first(). This is to be used for the case where the warning is
only executed in pass_first() code; this is highly discouraged as it
means the warnings will not appear in the list file and subsequent
passes may make the warning suddenly vanish.
ERR_PASS2 just as before suppresses an error or warning unless we are
in pass_final().
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
? in identifiers turns out to be used in the field even in non-TASM
mode. Resolve this by allowing it in an identifier still, but treat
'?' by itself the same as we would a keyword, meaning that it needs to
be separated from other identifier characters.
In other words:
a ? b : c ; conditional expression
a?b:c ; seg:off expression seg = a?b, off = c
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
New macro which defines the offset on an object rather than a
type. This macro, as far as I know, ought to be fully portable, unlike
the fallback version of offsetof().
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Make strlist_free() take a pointer to a pointer, so we can set it to
NULL.
Buffer warnings on a strlist until we either get an error or we are in
pass 2. Hopefully this should let us get rid of a lot of the ERR_PASS*
bullshit, which far too often causes messages to get lost.
asm/labels.c contains one example of a warning that cannot be made
correct with a specific pass number.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Make it a selectable option at allocation time if a strlist should
contain only unique strings or not. If not, we omit the hash table and
strlist_find() will not do anything.
Add printf()-style functions to a strlist.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Add a set of variants on the asprintf functions, "axprintf", which
allocate extra storage for metadata at the head of the allocated
buffer.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
It is extremely desirable to allow the user fine-grained control of
warnings, but this has been complicated by the fact that a warning
class has had to be defined in no less than three places (error.h,
error.c, nasmdoc.src) before it can be used in source code. Instead,
use a script to define these via magic comments at the point of use.
This hopefully will encourage creating new classes as needed.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
It is possible on memory exhaustion that nasm_fatal() might cause
another allocation error, thus calling nasm_alloc_failed() again. If
we find us in nasm_alloc_failed() for a second time, try to get a
message out and then call abort().
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
The prefix ERR_WARN_ is unnecessarily long and may be a disincentive
to create new warning categories. Change it to WARN_*, it is still
plenty distinctive.
This is equivalent to nasm-2.14.xx checkin 77f53ba6d4.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
The prefix ERR_WARN_ is unnecessarily long and may be a disincentive
to create new warning categories. Change it to WARN_*, it is still
plenty distinctive.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Add a version of (v)asprintf(), which allocates a string on the
heap. Unlike the standard version of (v)asprintf(), we return the
pointer; if one wants the length of the string then one can simply use
the %n pattern.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
The currently-unused strtbl was basically a slightly different version
of strlist, with the find and linearize capabilities. Merge these two
together by augmenting strlist to have the same capabilities.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Add binary key support to the hash table interface. Clean up the
interface to contain less extraneous crud.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Create our own ctype table where we can do the tests we want to do
cheaply, instead of calling ctype functions and then adding additional
tests all over the code.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
When we are running regression tests we compare binary
forms and the strings better to be the constants to not
trigger false positives.
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
Use a hash table to enforce uniqueness in a string list. It is still
an ordered list, however, and can be walked in insertion order.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Nearly all instances of nasm_fatal() and nasm_panic() take a flags
argument of zero. Simplify the code by making nasm_fatal and
nasm_panic default to no flags, and add an alternate version if flags
really are desired. This also means that every call site doesn't have
to initialize a zero argument.
Furthermore, ERR_NOFILE is now often not necessary, as the error code
will no longer cause a null reference if there is no current
file. Therefore, we can remove many instances of ERR_NOFILE which only
deprives the user of information.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Make the RAA infrastructure a bit cleaner, make it support 64-bit
indicies, and reduce the memory overhead of a sparse or small RAA --
the old code would allocate a *minimum* of 256K for each RAA. The new
code reduces that to 16K, and will not mandatorily allocate an entry
in the zero position.
The new shift, 11, was chosen so that a 32-bit RAA value will need 3
layers and a 64-bit value 6 layers, without excessive waste.
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
RAAs can, and should be, usable for storing pointers as well as
integers. In reality it is exactly the same code, but make it explicit
by having different entry points. In the actual RAA the data is
stored as a union, which in practice will not occupy any more space
than the existing code.
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Make any "deadman"-style execution limit configurable on the command
line (--limit-foo) or via a pragma (%pragma limit foo).
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Simplify the nasm_malloc() code by moving the pointer check into a
common subroutine.
We can now issue a filename error even for failures like malloc().
Add support for the gcc sentinel attribute (verify that a list ends
with NULL).
Add a handful of safe_alloc attributes.
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Get rid of remaining dependencies on FILENAME_MAX, which ought to have
been removed a long time ago.
Remove ofmt->filename(); all implementations pretty much do the same
thing and there is absolutely no reason to duplicate that
functionality all over the place.
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
