Re: [PATCH] Revert to defining __extern_inline only for g++-4.3+

["Carlos O'Donell" <carlos at redhat dot com>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 08/13/2014 02:23 PM, Siddhesh Poyarekar wrote:
> Hi,
> 
> The check for only __GNUC_STDC_INLINE__ and __GNUC_GNU_INLINE__ may
> not be sufficient since those flags were added during initial support
> for C99 inlining semantics.  There is also a problem with always
> defining __extern_inline and __extern_always_inline, since it enables
> inline wrapper functions even when GNU inlining semantics are not
> guaranteed.  This, along with the possibility of such wrappers using
> redirection (btowc for example) could result in compiler generating an
> infinitely recusrive call to the function.
> 
> In fact it was such a recursion that led to this code being written
> the way it was; see:
> 
> https://bugzilla.redhat.com/show_bug.cgi?id=186410
> 
> The initial change:
> 
> commit b7bfe116e6304da848759b69a6d713da3e93e936
> Author: Marek Polacek <polacek@redhat.com>
> Date:   Wed Sep 26 12:58:36 2012 +0200
> 
>     Fix up definitions for older compilers.
> 
> was to fix bugs 14530 and 13741, but they can be resolved by checking
> if __fortify_function and/or __extern_always_inline are defined, as it
> has been done in this patch.
> 
> I have tested this with gcc 3.2, 3.4 and 4.1.  In addition, I have
> done a quick audit of uses of __extern_always_inline and
> __extern_inline to make sure that none of the uses can result in
> compilation errors.
> 
> There is however a regression in this patch for llvm, since it reverts
> the llvm expectation that __GNUC_STDC_INLINE__ or __GNUC_GNU_INLINE__
> definition imply proper extern inline semantics.
> 
> If we don't want this fixed in 2.20 at this stage, I'd like to know if
> we can backport to the 2.20 branch after release.

I think this *should* be backported to 2.20.

OK for master with one comment expansion and [BZ #17266] in ChangeLog.

> 2014-08-13  Siddhesh Poyarekar  <siddhesh@redhat.com>
> 	    Jakub Jelinek  <jakub@redhat.com>
> 
> 	* libio/stdio.h: Check definition of __fortify_function
> 	instead of __extern_always_inline to include bits/stdio2.h.
> 	* math/bits/math-finite.h [__USE_XOPEN || __USE_ISOC99]: Also
> 	check if __extern_always_inline is defined.
> 	[__USE_MISC || __USE_XOPEN]: Likewise.
> 	[__USE_ISOC99] Likewise.
> 	* misc/sys/cdefs.h (__fortify_function): Define only if
> 	__extern_always_inline is defined.
> 	[!__cplusplus || __GNUC_PREREQ (4,3)]: Revert to defining
> 	__extern_always_inline and __extern_inline only for g++-4.3
> 	and newer or a compatible gcc.
> 
> diff --git a/libio/stdio.h b/libio/stdio.h
> index d8c0bdb..1f4f837 100644
> --- a/libio/stdio.h
> +++ b/libio/stdio.h
> @@ -932,7 +932,7 @@ extern void funlockfile (FILE *__stream) __THROW;
>  #ifdef __USE_EXTERN_INLINES
>  # include <bits/stdio.h>
>  #endif
> -#if __USE_FORTIFY_LEVEL > 0 && defined __extern_always_inline
> +#if __USE_FORTIFY_LEVEL > 0 && defined __fortify_function

OK.

>  # include <bits/stdio2.h>
>  #endif
>  #ifdef __LDBL_COMPAT
> diff --git a/math/bits/math-finite.h b/math/bits/math-finite.h
> index aa755de..0656645 100644
> --- a/math/bits/math-finite.h
> +++ b/math/bits/math-finite.h
> @@ -251,7 +251,8 @@ extern long double __REDIRECT_NTH (lgammal_r, (long double, int *),
>  # endif
>  #endif
>  
> -#if defined __USE_XOPEN || defined __USE_ISOC99
> +#if ((defined __USE_XOPEN || defined __USE_ISOC99) \
> +     && defined __extern_always_inline)

OK.

>  /* lgamma.  */
>  __extern_always_inline double __NTH (lgamma (double __d))
>  {
> @@ -284,7 +285,8 @@ __extern_always_inline long double __NTH (lgammal (long double __d))
>  # endif
>  #endif
>  
> -#if defined __USE_MISC || defined __USE_XOPEN
> +#if ((defined __USE_MISC || defined __USE_XOPEN) \
> +     && defined __extern_always_inline)

OK.

>  /* gamma.  */
>  __extern_always_inline double __NTH (gamma (double __d))
>  {
> @@ -422,7 +424,7 @@ extern long double __REDIRECT_NTH (sqrtl, (long double), __sqrtl_finite);
>  # endif
>  #endif
>  
> -#ifdef __USE_ISOC99
> +#if defined __USE_ISOC99 && defined __extern_always_inline

OK.

>  /* tgamma.  */
>  extern double __gamma_r_finite (double, int *);
>  __extern_always_inline double __NTH (tgamma (double __d))
> diff --git a/misc/sys/cdefs.h b/misc/sys/cdefs.h
> index 04db956..d8ee73c 100644
> --- a/misc/sys/cdefs.h
> +++ b/misc/sys/cdefs.h
> @@ -131,7 +131,6 @@
>  /* Fortify support.  */
>  #define __bos(ptr) __builtin_object_size (ptr, __USE_FORTIFY_LEVEL > 1)
>  #define __bos0(ptr) __builtin_object_size (ptr, 0)
> -#define __fortify_function __extern_always_inline __attribute_artificial__

OK.

>  
>  #if __GNUC_PREREQ (4,3)
>  # define __warndecl(name, msg) \
> @@ -318,12 +317,10 @@
>  # define __attribute_artificial__ /* Ignore */
>  #endif
>  
> -#ifdef __GNUC__
> -/* One of these will be defined if the __gnu_inline__ attribute is
> -   available.  In C++, __GNUC_GNU_INLINE__ will be defined even though
> -   __inline does not use the GNU inlining rules.  If neither macro is
> -   defined, this version of GCC only supports GNU inline semantics. */
> -# if defined __GNUC_STDC_INLINE__ || defined __GNUC_GNU_INLINE__
> +/* GCC 4.3 and above with -std=c99 or -std=gnu99 implements ISO C99
> +   inline semantics, unless -fgnu89-inline is used.  */

Please expand this comment to say that using just __GNUC_STDC_INLINE__ and 
__GNUC_GNU_INLINE__ is not a correct solution.

> +#if !defined __cplusplus || __GNUC_PREREQ (4,3)
> +# if defined __GNUC_STDC_INLINE__ || defined __cplusplus

OK.

>  #  define __extern_inline extern __inline __attribute__ ((__gnu_inline__))
>  #  define __extern_always_inline \
>    extern __always_inline __attribute__ ((__gnu_inline__))
> @@ -331,9 +328,10 @@
>  #  define __extern_inline extern __inline
>  #  define __extern_always_inline extern __always_inline
>  # endif
> -#else /* Not GCC.  */
> -# define __extern_inline  /* Ignore */
> -# define __extern_always_inline /* Ignore */
> +#endif
> +
> +#ifdef __extern_always_inline
> +# define __fortify_function __extern_always_inline __attribute_artificial__

OK.

>  #endif
>  
>  /* GCC 4.3 and above allow passing all anonymous arguments of an
> 

Marek's patch was a good attempt at cleaning up the semantics of the rules
that the glibc headers apply, but it reinforces that the rules are hard to
remember and keep straight without good documentation and continued testing
with older compiler.

Cheers,
Carlos.

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1

iQEcBAEBAgAGBQJUF6WfAAoJECXvCkNsKkr/PxwIANOHzbcr3aPEKBZRnLSe0mke
ogZtM5cjvHuPwhStaxqeX5sS+7BeWQ5GSNN4hZ/JrPHNao6R1hlQBfP6Oaf52Wbb
vru2PN9nuVKlgT4l9txSoKE7EB5VMBidTDqJ6ahFTgP1MAOZwncgOXXyRNZQ4sCG
qpjs3Yq1n0ImSZ5HcNckyW4YYhe5s6NMgz6yEOZf2fFppQgIUCq0q0XjY6VQsK/m
WS7loSeowz0A5T3y67WlNKMAKHIjekt7d8zlMQciZlM8+m+Fo28Hp/1cT0799yuZ
SMFbkmxDeDMI6xHkphS2hF6HeYh5Y6of1Wch38BPGCifdIpTr9Y72Ibo+BxrObI=
=tpsG
-----END PGP SIGNATURE-----