[tree-ssa] Fix for failure to build glibc

[Joseph S. Myers]

On Thu, 8 May 2003 law@redhat.com wrote:

> Understood.  But the reality is that this construct is clearly used and
> having tree-ssa break it won't fly.
> 
> If we're going to declare this code invalid, then we need to get the
> mainline compiler doing it first.

Making the compiler reject the code is something that needs to be done
deliberately, on mainline (though that doesn't mean anyone will actually
get round to doing it before tree-ssa is merged to mainline).  But when
it's rejected I don't see the need for any deprecation period - such code
has never had a sensible definition for its meaning, even if in the
particular case in use it happens to work through luck - and glibc ought
to be fixed not to jump out of statement expressions.  This is a case
where the semantics are so ill-defined that permitting such jumps
long-term for the sake of existing code that uses them (as might be done
for some unfortunate extensions with better-defined semantics, and actual
users) doesn't seem reasonable.

-- 
Joseph S. Myers
jsm28@cam.ac.uk