Paul Eggert <[email protected]> writes:
> That glibc patch needs more work, as I think Gnulib is misusing _LIBC.
> _LIBC means we're using the header while compiling glibc itself,
> whereas obstack.h appears in /usr/include/obstack.h and is used in
> other contexts.
I see.
In lib/cdefs.h there is '#ifndef __GNULIB_CDEFS' and in lib/glob.in.h
there is '# define __GLOB_GNULIB 1'.
Not the most elegant thing in the world, but perhaps gnulib-common.m4
could '#define __GNULIB 1' in config.h and that can be used. That is the
best idea that comes to mind at the moment.
The other idea I had was using _GL_ATTRIBUTE_PURE to detect Gnulib. Both
glibc and Gnulib have:
#ifndef __attribute_pure__
# define __attribute_pure__ _GL_ATTRIBUTE_PURE
#endif
So you could do:
#ifndef _GL_ATTRIBUTE_PURE
# define __SMALL_PTRDIFF_T 0
#endif
Since we #warn if config.h is not included before obstack.h and other
Gnulib replacement headers, this shouldn't cause any ambigous compiler
errors due to missing the definition.
Collin
