[Bug bootstrap/54659] [4.8 Regression] Bootstrap with --disable-nls broken under Windows

["rguenther at suse dot de" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54659

--- Comment #2 from rguenther at suse dot de <rguenther at suse dot de> 2012-10-26 12:05:14 UTC ---
On Fri, 26 Oct 2012, burnus at gcc dot gnu.org wrote:

> 
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54659
> 
> Tobias Burnus <burnus at gcc dot gnu.org> changed:
> 
>            What    |Removed                     |Added
> ----------------------------------------------------------------------------
>                  CC|                            |burnus at gcc dot gnu.org,
>                    |                            |rguenther at suse dot de
> 
> --- Comment #1 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-10-26 10:36:30 UTC ---
> From PR 54281:
> 
> | With --disable-nls intl.h does
> |
> | #ifdef ENABLE_NLS
> | #include <libintl.h>
> | extern void gcc_init_libintl (void);
> | extern size_t gcc_gettext_width (const char *);
> | #else
> | /* Stubs.  */
> | # undef textdomain
> | # define textdomain(domain) (domain)
> | # undef bindtextdomain
> | # define bindtextdomain(domain, directory) (domain)
> | # undef gettext
> | # define gettext(msgid) (msgid)
> |
> | which wrecks an included libintl.h:
> | We end up including libintl.h through
> |
> | gcc/double-int.h
> | #include <gmp.h>
> |   #include <iosfwd>  (here from GCC 4.1)
> |    #include <bits/c++locale.h>
> |      #include <libintl.h>
> 
> Side note: On my system only GCC 4.1's c++locale.h includes libintl.h, GCC
> 4.3/4.4/4.5/4.6/4.7 don't.
> 
> And:
> | Possibly by looking for and
> | including libintl.h before re-defining those macros?  It's the toplevel
> | intl.h btw.
> 
> | Another fix is to include all system headers (and thus gmp.h) from system.h
> | which always comes before includes of intl.h.
> 
> 
> 
> The patch for PR 54281 comment 10 (Rev. 190487) changed it to:
> 
>      #if defined(HAVE_LIBINTL_H) || defined(ENABLE_NLS)
>      # include <libintl.h>
>      #endif
> 
>      #ifdef ENABLE_NLS
> ...
> #else
>      /* Stubs. */
>      # undef textdomain
>      # define textdomain(domain) (domain)
> ...
> 
> 
> 
> As written in comment 0, that breaks MinGW builds as libintl.h re-defined
> fprint on non-POSIX-printf systems. If one now tries to undo the effect by
> using:
> 
> #if defined(HAVE_LIBINTL_H) || defined(ENABLE_NLS)
> #  include <libintl.h>
> #  if !defined(ENABLE_NLS)
> #    undef fprintf
> #    undef sprintf
> #    undef snprintf
> #    undef vfprintf
> #  endif
> #endif
> 
> 
> One ends up with:
> 
> libbackend.a(ipa-pure-const.o): In function `check_decl':
> C:\MinGW\msys\1.0\home\brad\gfortran\ibin\gcc/../../gcc-trunk/gcc/ipa-pure-const.c:253:
> undefined reference to `_libintl
> _fprintf'
> ...
> libcommon-target.a(opts.o): In function `wrap_help':
> C:\MinGW\msys\1.0\home\brad\gfortran\ibin\gcc/../../gcc-trunk/gcc/opts.c:884:
> undefined reference to `___printf__'
> 
> 
> 
> Without having tested with neither MinGW nor with GCC 4.1/Linux, I wonder
> whether one could do something like:
> 
>      #if defined(HAVE_LIBINTL_H) && !defined(ENABLE_NLS)
>         # define _LIBINTL_H 1
>         #end if
> 
>         #ifdef (ENABLE_NLS)
>      # include <libintl.h>
>      #endif
> 
> Or is that too evil or won't it work?

Fact is that all this stuff happens because gmp.h is not included
from system.h ...