On Thu, 22 May 2003 15:24:06 +0200, Carl-Daniel Hailfinger said:
> Nobody has found an error in the code we talked about, so a compiler bug
> in gcc 3.2.3 seems to be the only explanation.
In the last 20 years, I've come across lots of cases where optimizers did
foolish things that broke code. I've also come across the odd case or three
where the optimizer merely exposed a bug. Favorite cases here are where
the optimizer removes what it thinks is a dead/redundant load/store, and
exposes a race condition on a variable that should have been 'volatile' but
wasn't, odd corner cases where sequence points actually matter (one of these was
just posted here the other day, in fact)... stuff like that.
So yes. It's probably something borked in gcc 3.2.3 - but we probably won't
know for sure till somebody goes over the assembler output with a fine tooth
comb...
This archive was generated by hypermail 2b29 : Fri May 23 2003 - 22:00:49 EST