GCC 3.2 Branch Closure

[Joe Buck]

> I believe that HJ's patch for x86 clears up the last of the significant
> silent code generation problems, and Joe is right that we have fixed a
> ton of regressions in GCC 3.2.1; we should get that compiler out there
> in the wild, and start focusing on GCC 3.3, and GCC 3.2.2 if necessary.

No, there's still 5351/7951.

HJ has two different patches; the one that fixes 8146 has been applied,
but the other one, that fixes 5351 and 7951, has not been.  I think that
the other one, in

http://gcc.gnu.org/ml/gcc-patches/2002-08/msg01170.html

should be applied.  I had raised an alarm about a new regression
introduced by the latter patch, but a subsequent, more careful test
convinced me that I was wrong: I am seeing no regressions (in fact,
no FAILs of any kind) with the patch as applied to today's branch.