Re: AC_CHECK_DECL has wrong results on macOS
Bruno Haible via Bug reports for autoconf <[email protected]> Mon, 26 May 2025 11:18:53 +0200
| Newsgroups | gmane.comp.sysutils.autoconf.bugs |
|---|---|
| Organization | GNU |
| Message-ID | <5898890.tIAgqjz4sF@nimes> |
Hi Paul,
> > Here's a proposed fix for the issue along these lines.
>
> Thanks. It'd be nicer if AC_CHECK_DECL didn't require AC_CANONICAL_HOST.
> We can do that by checking inside the C program rather than using
> AC_CANONICAL_HOST.
This comes with the cost of an extra compiler invocation on all non-macOS
systems. But I understand the desire to avoid AC_CANONICAL_HOST.
> Also, shouldn't we simply test for the option in question rather than
> first checking for Clang? GCC might eventually add it. This way, we
> don't need to test for Clang.
True. I was worried that other compilers would accept said option and
then warn about it at each compilation (like we have seen with some other
compiler option, I don't remember which one). But I now tested with
AIX xlc and Sun cc, and my worry was unfounded.
> So, how about the attached patch instead?
Looks good and works fine. Tested with the above-mentioned minos.tar.gz
- on the cfarm104.cfarm.net machine,
- with AIX xlc and Sun cc: The configure script reports e.g.
checking for cc -D_STDC_C99= -xarch=generic64 -O options to detect undeclared functions... none needed
checking for cc -D_STDC_C99= -xarch=generic64 -O options to ignore future-version functions... none needed
Since you made significant changes to my input, please can you mark yourself
as co-author in the git commit message? [1]
Co-authored-by: Paul Eggert <[email protected]>
Thanks.
Bruno
[1] https://stackoverflow.com/questions/58525836/