Re: AIX 4.3.3 IBM C compiler patch for m4-1.4.8

[Albert Chin]

On Sun, Feb 04, 2007 at 07:52:08AM -0700, Eric Blake wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> According to Albert Chin on 2/4/2007 7:47 AM:
> > The IBM C compiler on  AIX 4.3.3 (v6) doesn't like signed bit types:
> >   gmake[1]: Entering directory `/opt/build/m4-1.4.8/src'
> >   xlc    -I../lib -I../lib   -O2 -qro -qroconst -qmaxmem=-1 -qarch=com -c 
> > m4.c
> >   "m4.h", line 383.3: 1506-009 (S) Bit-field gnu_extension must be of type 
> > signed int, unsigned int or int.
> >   "m4.h", line 384.3: 1506-009 (S) Bit-field groks_macro_args must be of 
> > type signed int, unsigned int or int.
> >   "m4.h", line 385.3: 1506-009 (S) Bit-field blind_if_no_args must be of 
> > type signed int, unsigned int or int.
> >   gmake[1]: *** [m4.o] Error 1
> > 
> > Patch attached.
> 
> > -  bool traced : 1;
> > +  unsigned int traced : 1;
> 
> Thanks for the report.  However, I would tend to think that this is a bug
> in the gnulib stdbool module.  Is ./configure detecting that the AIX bool
> is sufficient, or is it using the gnulib replacement?  If it is getting
> the AIX bool, then the gnulib check should be enhanced to check that bool
> can be used for bitfields.  If it is getting the gnulib bool, then we
> should figure out how to make the gnulib bool pick a type that will work
> for this usage pattern.

$ grep -i bool lib/config.h
/* Define to 1 if stdbool.h conforms to C99. */
/* #undef HAVE_STDBOOL_H */
/* Define to 1 if the system has the type `_Bool'. */
/* #undef HAVE__BOOL */

lib/stdbool.h is being used and _Bool is defined to 'signed char':
  #  define _Bool signed char

I thought about it being the culprit but it really isn't. It's simply
in using the bool type as a bit type. So, I don't see the need to
penalize stdbool.

-- 
albert chin (address@hidden)