Re: Gnulib's alloca.h used even when there is a system header

[Eli Zaretskii]

> Cc: bug-gnulib@gnu.org
> From: Paul Eggert <egg...@cs.ucla.edu>
> Date: Sun, 17 Feb 2019 13:09:06 -0800
> 
> Eli Zaretskii wrote:
> >   #ifndef alloca
> >   # ifdef __GNUC__
> > -#  define alloca __builtin_alloca
> > +#  if HAVE_ALLOCA_H
> > +#   include_next <alloca.h>
> > +#  else
> > +#   define alloca __builtin_alloca
> > +#  endif
> >   # elif defined _AIX
> 
> Why do the 'include_next' only for GCC? Why not do it for all compilers?

I didn't know other compilers supported include_next.

> Also, what platforms don't work with the current Gnulib alloca module, and 
> why?

I bumped into this with mingw.org's MinGW, which introduced alloca.h
in its recent versions.  (MinGW64 doesn't have this header.)  This new
MinGW alloca.h does this:

  __CRT_ALIAS void *alloca( size_t __n ){ return __builtin_alloca( __n ); }

which makes alloca an always-inline function by virtue of

  #  define __CRT_INLINE   extern __inline__
  #  define __CRT_ALIAS  __CRT_INLINE __attribute__((__always_inline__))

The other part of the puzzle is that stdlib.h does this:

  # include "alloca.h"

so Gnulib's alloca.h is bypassed.  Which causes this when compiling
Gnulib's vasnprintf.c:

     In file included from vasnprintf.c:57:0:
     d:\usr\include\alloca.h: In function 'vasnprintf':
     ./alloca.h:40:18: error: inlining failed in call to always_inline 
'__builtin_alloca': function not considered for inlining
      #  define alloca __builtin_alloca
                       ^
     In file included from d:\usr\include\stdlib.h:499:0,
                      from vasnprintf.c:71:
     d:\usr\include\alloca.h:62:48: note: called from here
      __CRT_ALIAS void *alloca( size_t __n ){ return __builtin_alloca( __n ); }
                                                     ^~~~~~~~~~~~~~~~~~~~~~~
     Makefile:945: recipe for target `vasnprintf.lo' failed