This is the mail archive of the
mailing list for the GCC project.
Re: [3.3 branch] IA64 bootstrap failure
email@example.com (H. J. Lu) wrote on 09.07.03 in <20030709152453.GA3682@lucon.org>:
> On Tue, Jul 08, 2003 at 03:53:38PM -0700, Jim Wilson wrote:
> > Andreas Schwab wrote:
> > >Running the same (stage2) compiler command on the command line produces
> > >output that is the same as the output from the stage1 compiler. So it
> > >indeed looks like it's dependent on the environment size.
> > I've debugged problems like this before. You vary the environment size
> > until you can reproduce the problem, and then you debug two compilers in
> > parallel with different environment sizes until you figure out why they
> > emit different code. Probably an uninitialized variable or a
> > stray/out-of-bounds memory reference somewhere. Another possibly is
> > that addresses are being used (e.g. in a hash table) and there is a
> > latent bug in the code.
> A few things I noticed:
> 1. It doesn't happen with -da.
> 2. It doesn't happen with debug is off.
> 3. It doesn't happen with preprocessed input.
> 3. It is very sensitive to input. Removing a few lines or adding
> "#if 0/#endif" will make it to go away.
Which is of course entirely consistent with the hypothesis that it relies
on addresses somehow.