PR 8146

[Joe Buck]

> I wrote:
> > Anyway, I've now started a 2.95.3 build using 3.2 as the host
> > compiler, without -fno-strict-aliasing, and without -O2 (really,
> > this time) for stage 1, but with -O2 in BOOT_CFLAGS (which shouldn't
> > be a problem, right?).
> 
> This still fails for me.
> Bootstrap comparison failure!
> cccp.o differs
> f/data.o differs
> f/target.o differs

Can you confirm, by looking at the output from make bootstrap (you saved
it, right?), that gcc 3.2 is never given any -O option of any kind?
(I could try this myself, but if you already have it ...).

The key thing we need to understand is whether this is a violation of the
strict-aliasing rules in the 2.95.3 source, or a 3.2 bug.

I don't want to hear about values of makefile variables, I want to see
the actual gcc 3.2 commands that are run.