[PATCH] Fix libiberty.h inclusion on non-glibc hosts tree-wide
Sam James
sam@gentoo.org
Wed Nov 18 12:12:31 GMT 2026
Nick Alcock <nix@esperi.org.uk> writes:
> On 18 Nov 2026, Sam James stated:
>
>> Nick Alcock <nix@esperi.org.uk> writes:
>>
>>> In 2025, Iain Sandoe added memrchr support to libiberty.h. This requires
>>> a new AC_CHECK_FUNCS of memrchr in every project that transitively
>>> includes it, or builds on non-glibc, non-Linux hosts will fail (observed
>>> on FreeBSD 15).
>>
>> That doesn't sound right to me as stated: it would mean nobody has reported or
>> tested that in over a year, even for say musl? Or do you mean 'non-glibc
>> and non-Linux'? (I guess you do based on the condition in libiberty.h
>> now I look).
>>
>> Anyway, does it really require that?
>>
>> include/libiberty.h has:
>>
>> /* Make memrchr available on systems that do not have it. */
>> #if !defined (__GNU_LIBRARY__ ) && !defined (__linux__) && \
>> !defined (HAVE_MEMRCHR)
>> extern void *memrchr(const void *, int, size_t);
>> #endif
>>
>> so on say, FreeBSD, I would expect this to be used. Then you get the
>> link-time definition from libiberty?
>
> Like glibc, FreeBSD these days defines memrchr but uses extra macro
> magic which disagrees with a redefinition of the prototype unless the
> prototype is in scope. So if you don't check for it in the subproject's
> configury, all those tests fail and we hit the prototype and then
> explode.
>
> Maybe the right fix is to also add a #ifndef memrchr outside all the
> others, on the grounds that if the system already defines memrchr as a
> macro we are certain to not need to redefine it? The same might be worth
> doing to the declarations of the other POSIX-specified things in
> libiberty.h (of which there are a quite a few, and quite a few
> AC_CHECK_FUNCS of them in projects that don't use them just to stop
> things breaking at #include time).
This is what I was going to suggest, but...
>
>> Can you show me what the failure is please?
>
> CC targets.lo
> In file included from /usr/include/stdio.h:38,
> from ../../bfd/sysdep.h:30,
> from ../../bfd/targets.c:22:
> ../../bfd/../include/libiberty.h:226:14: error: expected identifier or '(' before '_Generic'
> 226 | extern void *memrchr(const void *, int, size_t);
> | ^~~~~~~
> gmake[3]: *** [Makefile:2329: targets.lo] Error 1
> gmake[3]: *** Waiting for unfinished jobs....
>
> (and so on for every other affected file.)
>
> ... because on FreeBSD, memrchr is a macro:
>
> #define memrchr(b, c, n) __qualsel((b), \
> (const void *)(memrchr)((b), (c), (n)), (memrchr)((b), (c), (n)))
>
> (where __qualsel is a _Generic usage.)
.. now you show this, I realise why it sounded familiar:
https://gcc.gnu.org/PR127914.
That is, we should just do (memrchr) instead.
>
>>> Fix trivial, but a tad invasive, since a lot of projects include
>>> libiberty.h, and it shows up a few places that don't include config.h at
>>> all (including some that include ansidecl.h, which requires config.h for
>>> proper operation).
>>>
>>> All relevant configury regenerated.
>>>
>>> Also touches GDB machinery, so gdb-patches is Cc:ed.
>>>
>>> Iain Sandoe Cc:ed: Iain, if you disagree with my rather nasty approach
>>> here, I'd be interested in anything better. (The need to use %top in
>>> binutils/arlex.l is slightly worrying. It's not in POSIX, but
>>> we need something like it or we can't include config.h before the
>>> system headers it must precede. (Including config.h in libiberty.h
>>> won't work for the same reason.)
>>>
>>> But if you don't do this, you can't compile binutils/arlex.c on
>>> FreeBSD any more (or any non-Linux, non-glibc system).
>>
>> Sorry if I'm missing something obvious here, it's a busy fortnight
>> for me but I wanted to reply before we do something that may be the
>> wrong direction.
>
> I spent the entire time I was working on this commit thinking I was
> missing something obvious. Hence the need for review :)
:)
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 418 bytes
Desc: not available
URL: <https://sourceware.org/pipermail/binutils/attachments/20261118/10787ee3/attachment.sig>
More information about the Binutils
mailing list