[PATCH v4] assert: Refactor assert/assert_perror

[Adhemerval Zanella Netto]

On 26/09/25 07:44, Yury Khrustalev wrote:
> Hi Adhemerval,
> 
> On Thu, Aug 21, 2025 at 08:48:13AM -0300, Adhemerval Zanella wrote:
>> It now calls __libc_assert, which contains similar logic. The assert
>> call only requires memory allocation for the message translation, so
>> test-assert2.c is adapted to handle it.
>>
>> It also removes the fxprintf from assert/assert_perror; although it
>> is not 100% backwards-compatible (write message only if there is a
>> file descriptor associated with the stderr). It now writes bytes
>> directly without going through the wide stream state.
>>
>> Checked on aarch64-linux-gnu.
> 
> This change causes compilation error when compiling with -Og:
> 
> In file included from ../include/sys/cdefs.h:10,
>                  from ../include/features.h:524,
>                  from ../stdlib/errno.h:25,
>                  from ../include/errno.h:2,
>                  from ../sysdeps/unix/sysv/linux/libc_fatal.c:19:
> ../include/stdio.h: In function ‘__libc_message_wrapper’:
> ../misc/sys/cdefs.h:639:30: error: invalid use of ‘__builtin_va_arg_pack_len ()’
>   639 | # define __va_arg_pack_len() __builtin_va_arg_pack_len ()
>       |                              ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
> ../include/stdio.h:191:7: note: in expansion of macro ‘__va_arg_pack_len’
>   191 |   if (__va_arg_pack_len () > LIBC_MESSAGE_MAX_ARGS)
>       |       ^~~~~~~~~~~~~~~~~
> In file included from ../sysdeps/posix/libc_fatal.c:23,
>                  from ../sysdeps/unix/sysv/linux/libc_fatal.c:36:
> ../include/stdio.h:194:7: error: call to ‘__libc_message_error’ declared with attribute error: invalid number of arguments
>   194 |       __libc_message_error ();
>       |       ^~~~~~~~~~~~~~~~~~~~~~~
> ../include/stdio.h:197:3: error: invalid use of ‘__builtin_va_arg_pack ()’
>   197 |   __libc_message_impl (vmaname, fmt, __va_arg_pack ());
> 
>> ...
>>

Right, I haven't see it because this is not an usual compiler option.

>> diff --git a/include/stdio.h b/include/stdio.h
>> index 3d917dba5b..d9a839e115 100644
>> --- a/include/stdio.h
>> +++ b/include/stdio.h
>>
>> ...
>>
>> +static inline _Noreturn void __libc_message_wrapper (const char *vmaname,
>> +						     const char *fmt, ...)
> 
> I think this should also have __attribute__ ((__always_inline__)) to
> make use of __builtin_va_arg_pack_len() valid for all optimisation
> levels.

It seems reasonable, although I think we will need to fix other issues
to fully enable -Og support.  With gcc 15.1 I am seeing some build issues
with -Werror:

malloc.c:3315:20: error: ‘csize’ may be used uninitialized [-Werror=maybe-uninitialized]
 3315 |           && csize == nb
      |              ~~~~~~^~~~~
malloc.c: In function ‘_mid_memalign’:
malloc.c:3297:14: note: ‘csize’ was declared here
 3297 |       size_t csize;
      |              ^~~~~
cc1: all warnings being treated as errors


regexec.c: In function ‘check_node_accept_bytes’:
regexec.c:3776:36: error: ‘extra’ may be used uninitialized [-Werror=maybe-uninitialized]
 3776 |               const unsigned char *coll_sym = extra + cset->coll_syms[i];
      |                                    ^~~~~~~~
regexec.c:3762:42: note: ‘extra’ was declared here
 3762 |           const unsigned char *weights, *extra;
      |                                          ^~~~~


res_send.c: In function ‘reopen’:
res_send.c:856:21: error: ‘slen’ may be used uninitialized [-Werror=maybe-uninitialized]
  856 |                 if (__connect (EXT (statp).nssocks[ns], nsap, slen) < 0) {
      |                     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
res_send.c:804:27: note: ‘slen’ was declared here
  804 |                 socklen_t slen;
      |                           ^~~~

These are issues that we already suppress for -Os, so I think we will need to
make it enable for -Og as well.