On Fri, Feb 19, 2016 at 11:08:48AM +0000, Bernd Edlinger wrote:
> On 19.02.2016 11:56, Jakub Jelinek wrote:
> >
> > On Fri, Feb 19, 2016 at 10:50:34AM +0000, Bernd Edlinger wrote:
> > > While I think that we should probably not define __GNUC_GNU_INLINE__ at
> > > all for C++,
> > > because it is meaningless, I am warned that this could break (already
> > > broken) header files.
> >
> > It is not meaningless. The various headers need to know if it is safe to
> > use the gnu_inline attribute in C++.
> >
> > In any case, the desirable state is that e.g. the -E -dD output should be
> > identical if you explicitly request the default -std= version vs. if it is
> > set implicitly. We should verify it is the case even for C.
> >
> > Jakub
>
> I absolutely agree with you.
> The correct solution is probably doing this:
>
> --- gcc/cp/cfns.h.jj 2016-01-04 15:30:50.000000000 +0100
> +++ gcc/cp/cfns.h 2016-02-19 12:00:15.730375049 +0100
> @@ -124,9 +124,6 @@
>
> #ifdef __GNUC__
> __inline
> -#ifdef __GNUC_STDC_INLINE__
> -__attribute__ ((__gnu_inline__))
> -#endif
> #endif
> const char *
> libc_name_p (register const char *str, register unsigned int len)
This is of course wrong. cfns.h is a generated header, so you shouldn't
patch it.
If it is regenerated with a newer gperf (I have 3.0.4 installed), you get there:
@@ -124,7 +124,7 @@ hash (register const char *str, register
#ifdef __GNUC__
__inline
-#ifdef __GNUC_STDC_INLINE__
+#if defined __GNUC_STDC_INLINE__ || defined __GNUC_GNU_INLINE__
__attribute__ ((__gnu_inline__))
#endif
#endif
Jakub
