GCC Warning on MIPS

[Ralf Corsepius]

On Tue, 2008-04-22 at 10:13 -0500, Joel Sherrill wrote:
> Ralf Corsepius wrote:
> > On Tue, 2008-04-22 at 10:01 -0500, Joel Sherrill wrote:
> >   
> >> Ralf Corsepius wrote:
> >>     
> >>> On Tue, 2008-04-22 at 09:25 -0500, Joel Sherrill wrote:
> >>>
> >>>       
> >>>> Ralf,
> >>>>
> >>>> I know neither of us go digging through build
> >>>> logs when the build works, but I came across this
> >>>> warning building GCC SVN trunk targeting mips-rtems4.9
> >>>>
> >>>> In file included from ./tm.h:10,
> >>>>                  from ../../gcc/gcc/gencheck.c:24:
> >>>> ../../gcc/gcc/config/rtems.h:21:1: warning: "NO_IMPLICIT_EXTERN_C" redefined
> >>>> In file included from ./tm.h:8,
> >>>>
> >>>> Not critical but is repeated about 10,000 times. :-D
> >>>>
> >>>> I think the #define NO_IMPLICIT_EXTERN_C in gcc/config/rtems.h
> >>>> needs an ifndef or an undef added.
> >>>>
> >>>> What do you think we need to do?
> >>>>
> >>>>         
> >>> Hmm, do you have this patch below applied?
> >>>
> >>> diff -Naur gcc-4.3.0.orig/gcc/config/mips/elf.h gcc-4.3.0/gcc/config/mips/elf.h
> >>> --- gcc-4.3.0.orig/gcc/config/mips/elf.h        2007-08-02 12:49:31.000000000 +0200
> >>> +++ gcc-4.3.0/gcc/config/mips/elf.h     2008-03-06 16:22:30.000000000 +0100
> >>> @@ -48,6 +48,4 @@
> >>>  #undef  ENDFILE_SPEC
> >>>  #define ENDFILE_SPEC "crtend%O%s crtn%O%s"
> >>>
> >>> -#define NO_IMPLICIT_EXTERN_C 1
> >>> -
> >>>  #define HANDLE_PRAGMA_PACK_PUSH_POP 1
> >>>
> >>>
> >>> This is a general bug in the mips port (I _know_ it is), but ...
> >>> upstream refuses to apply it, which forces us to keep it as a separate
> >>> patch :)
> >>>
> >>>
> >>>       
> >> In many cases throughout the gcc config files, they just undef
> >> the previous one and redefine.  In the above, ENDFILE_SPEC
> >> is done that way.  Three of the five definitions in config/rtems.h
> >> have an undef in front of them.  Would it be so bad to just
> >> do that for the rest of the ones in rtems.h?
> >>     
> >
> > Why should they?
> >
> > Wrt. to NO_IMPLICIT_EXTERN_C, to me this doesn't make any sense at all.
> >
> >   
> Maybe not but there is precedence for it.  The shared
> config files for a number of other OSes do this:
> 
> [joel at iceland config]$ grep -r NO_IMPLICIT_EXTERN_C .| grep ifndef | 
> grep -v .svn
> ./lynx.h:#ifndef NO_IMPLICIT_EXTERN_C
> [joel at iceland config]$ grep -r NO_IMPLICIT_EXTERN_C .| grep undef| grep 
> -v .svn
> ./netbsd.h:#undef  NO_IMPLICIT_EXTERN_C
> ./rs6000/sysv4.h:#undef  NO_IMPLICIT_EXTERN_C
> ./i386/sco5.h:#undef NO_IMPLICIT_EXTERN_C
> ./freebsd.h:#undef  NO_IMPLICIT_EXTERN_C
> ./vx-common.h:#undef  NO_IMPLICIT_EXTERN_C
> ./sparc/openbsd64.h:#undef NO_IMPLICIT_EXTERN_C

And where would be the sense behind this?

We want to force NO_IMPLICIT_EXTERN_C.

The bug is in mips/elf.h. This file belongs to the mips backend.

This file simply is broken by defining an OS specific flag such as
NO_IMPLICIT_EXTERN_C.

Cf. how we set up _all_ other targets. The only one behaving differently
is this broken mips file.

Ralf