Re: Compilation failure for libunistring-0.9.3 on ia64 hpux 11.23 and 11.31

[Gary V. Vaughan]

Hi Bruno,

Either I haven't been able to reach you at your clisp address, or your
responses have not reached me, so I'm reposting to bug-gnulib.

On Sun, Apr 17, 2011 at 11:18:44AM +0700, Gary V. Vaughan wrote:
> Ping?
> 
> On Wed, Apr 13, 2011 at 12:57:40AM +0700, Gary V. Vaughan wrote:
> > Not sure whether this is the gnulib stdbool tests gone awry or a bug
> > in libunistring itself... I suspect the latter because some files
> > include <stdbool.h> and get the system stdbool which defines bool as
> > _Bool, where others include "unistring/stdbool.h" which doesn't trust
> > the system bool declaration and sets it to unsigned char.
> > 
> > This results in a prototype mistmatch when compiling
> > lib/unicase/u16-is-invariant.c, which first includes unicase/invariant.h,
> > which itself includes <stdbool.h> so that the prototypes in
> > unicase/invariant.h are declared with bool expanded to _Bool.
> > 
> > Then a few lines further down, unicase/u16-is-invariant.c includes
> > "unistr.h", which includes "unistring/stdbool.h", and that changes the
> > bool definition so that when unicase/u-is-invariant.h is pulled in at
> > the end of the file, the function definitions now have bool expanded to
> > unsigned char, which the compiler flags as a type mismatch error.
> > 
> > I can force the compilation of that file through by passing -D_GL_STDBOOL_H
> > (which is obviously wrong, but might be a useful data point) to make
> > sure that compilation unit gets only the system bool definition... but then
> > I get a similar error on the next file.
> > 
> > cc -DHAVE_CONFIG_H -DNO_XMALLOC -I. -I..  -I. -I. -I.. -I.. 
> > -DIN_LIBUNISTRING -DDEPENDS_ON_LIBICONV=1 -I/opt/TWWfsw/libiconv113/include 
> > -z +O2 +Olit=all +Ofltacc +Oentrysched +Odataprefetch -c 
> > unicase/u16-is-invariant.c
> > "unicase/u-is-invariant.h", line 19: error #2147: declaration is 
> > incompatible
> >           with "int u16_is_invariant(const unistring_uint16_t *, size_t,
> >           unistring_uint16_t *(*)(const unistring_uint16_t *, size_t, const
> >           char *, uninorm_t, unistring_uint16_t *, size_t *), const char *,
> >           _Bool *)" (declared at line 33 of "./unicase/invariant.h")
> >   FUNC (const UNIT *s, size_t n,
> >   ^
> > 
> > 1 error detected in the compilation of "unicase/u16-is-invariont.c".
> > 
> > yukio% grep -i bool config.h
> > /* Define to 1 if stdbool.h conforms to C99. */
> > #define HAVE_STDBOOL_H 1
> > /* Define to 1 if the system has the type `_Bool'. */
> > #define HAVE__BOOL 1
> > 
> > Is this a known bug?  Can I pull a fix from somewhere?  Or is there
> > anything I can provide to help you track down the root cause of the bug?
> > 
> > [there was no bug address in configure.ac, so I took the liberty of
> >  contacting you directly.  Please forward this message to an appropriate
> >  mailing list and continue the thread there if you prefer, but please
> >  keep this address in the Cc:.]

I was able to solve this particular problem by replacing all references
to unistring/stdbool.h with stdbool.h so that the same definitions are
present everywhere.  Applying this change and recompiling even on Linux
doesn't affect the testsuite results, though I can't imagine that you
would have gone to the trouble of patching the gnulib files to include
a second stdbool.h if there weren't good reason... am I missing something,
or were my hosts perhaps not affected by whatever you were fixing with
the additional stdbool.h?

Also I have a couple of other small changes I had to make to compile
libunistring-0.9.3 on all the architectures I have access to.  Please
contact me off-list if you'd like the patches.

Cheers,
-- 
Gary V. Vaughan (gary AT gnu DOT org)

pgpLZFVyvGKaY.pgp
Description: PGP signature