Re: [PATCH] fix regression after linux.h / gnu-user.h split

["Joseph S. Myers" <joseph at codesourcery dot com>]

On Tue, 18 Jan 2011, Robert Millan wrote:

> 2011/1/12 Richard Guenther <richard.guenther@gmail.com>:
> > On Wed, Jan 12, 2011 at 1:44 PM, Joseph S. Myers
> > <joseph@codesourcery.com> wrote:
> >> Ping^2.  This patch
> >> <http://gcc.gnu.org/ml/gcc-patches/2010-12/msg02055.html> (with typo fix
> >> as noted in <http://gcc.gnu.org/ml/gcc-patches/2010-12/msg02056.html>) is
> >> pending review.
> >
> > Ok.
> 
> Sorry for noticing too late, but this patch introduced a small regression:
> TARGET_C99_FUNCTIONS and TARGET_HAS_SINCOS, which
> are always wanted on Glibc, are not defined anymore for GNU
> systems on kernels other than Linux.

You'll need to explain more why you think there is a regression.  Detail 
the before-and-after sets of macros defined on a specific target.

The key principle of this patch is that *targets use gnu-user.h if and 
only if they use linux.h*.  It should not change the set of macros defined 
for any target, because all targets that used linux.h continue to do so.  
Defining TARGET_C99_FUNCTIONS and TARGET_HAS_SINCOS in the correct 
target-dependent way is explicitly meant to be a subsequent, more risky, 
patch (and this would not be the same approach as in your patch, but a 
more complicated change involving config.gcc as well).

-- 
Joseph S. Myers
joseph@codesourcery.com