[PATCH] Fix PR bootstrap/26055

[Andrew Pinski]

> 
> Richard Guenther wrote:
> > This fixes bootstrap on darwin where including math.h conflicts with
> > defining __isinf.  This is done by a configure check for this exact
> > situation and disables libgcc-math in this case.
> > 
> > Checked by configuring with and without a modified math.h.
> > 
> > A better fix is in the works, but as this will restore bootstrap for
> > darwin, I'll commit it as obvious.
> 
> This is not obvious.  If the better fix is in the works, darwin people 
> may as well use --disable-libgcc-math (or make -k simply...).  Stuff has 
> been broken for a while last week on several platforms.

Actually libgomp only has one build bug left, using its own TLS detector.
libgcc-math caused more build failures on x86 non GNU/Linux targets
than any other patch I know of.

What I would do is stop and think for a moment and just remove libgcc-math
until these issues have been resolved.

I am only mad because there are ways of getting access to the targets, all
you have to do is ask.  NetBSD for an example can be found on the
CompileFarm: http://gcc.gnu.org/wiki/CompileFarm

Please next time test on more than just x86-linux-gnu when a patch enables it
for all of x86 and there is even a slight notation of it breaking other targets.

The other reason why I am upset is because I just got x86-darwin back to a working
state and the testsuite not looking that bad and this comes along and breaks it.  I
also was working on a patch to fix most of the rest of the failures also for x86-darwin.

-- Pinski