Ugh, I thought Clang had added support for this years ago. But it looks
like the change (https://reviews.llvm.org/D34158) never actually made it
in; it ran into some test failures after being committed and was
reverted, and then never reapplied. :(


On Fri, Apr 16, 2021 at 2:51 PM Rich Felker <dalias@libc.org> wrote:

> On Fri, Apr 16, 2021 at 02:57:21PM -0300, Érico Nogueira wrote:
> > Em 16/04/2021 11:26, Rich Felker escreveu:
> > >On Thu, Apr 15, 2021 at 09:35:21PM -0300, Érico Nogueira wrote:
> > >>GCC source code does contain a function to pre-include the
> > >><stdc-predef.h> header for glibc targets, but even so glibc still
> >
> > I seem to have been mistaken about the feature being glibc specific;
> using
> >
> >     echo "" | cc -xc - -E
> >
> > it seems the file does end up being included automatically.
> >
> > However, when using clang instead of gcc, it isn't included
> > automatically. I don't know if this is something that clang ought to
> > fix, is there some sort of standard about <stdc-predef.h>? Michael
> > Forney's cproc compiler doesn't seem to touch it either.
>
> It's not a standard, but given that it's established I don't see any
> reasonable argument for other compilers not to just do the same. You
> can always fix them manually with CC="clang -include stdc-predef.h" or
> similar though.
>
> > >>includes it in their own <features.h> header. furthermore, even if GCC
> > >>implemented this for musl targets, it is still necessary for other
> > >>compilers or previous versions of GCC.
> > >>---
> > >>  include/features.h | 2 ++
> > >>  1 file changed, 2 insertions(+)
> > >>
> > >>diff --git a/include/features.h b/include/features.h
> > >>index 85cfb72a..f3d53cbe 100644
> > >>--- a/include/features.h
> > >>+++ b/include/features.h
> > >>@@ -1,6 +1,8 @@
> > >>  #ifndef _FEATURES_H
> > >>  #define _FEATURES_H
> > >>+#include <stdc-predef.h>
> > >>+
> > >>  #if defined(_ALL_SOURCE) && !defined(_GNU_SOURCE)
> > >>  #define _GNU_SOURCE 1
> > >>  #endif
> > >>--
> > >>2.31.1
> > >
> > >I've hesitated to do this because features.h is not consistently
> > >included from all standard headers (only if it's needed), and the
> > >result would be inconsistent exposure of these macros. (Also
> > >inconsistent if they're checked before any standard headers are
> > >included, which is unfixable.) I think it makes more sense to just add
> > >"-include stdc-predef.h" to the compiler specfile or equivalent if it
> > >doesn't auto-include it, so that you get behavior that actually
> > >matches the spec.
> >
> > Do you know if clang can use the specfile? That would make it worth
> > it adding the entry, since GCC has the expected behavior already.
>
> No; specfiles are highly tied to GCC's compiler driver architecture.
> clang might have some other equivalent mechanism though.
>
> Rich
>