Segfaults building Rakudo

[Andrew Whitworth]

Moritz has been seeing segfaults during the Rakudo build/test/spectest
for the past several days. Bisect shows that they started at the
imcc_compreg_pmc merge. Other people have seen segfaults appear during
the Rakudo build, although they are always different for different
people. I've seen them under certain conditions on my machines, I have
also received reports of them from plobsing and benabik, maybe others.

plobsing and I each put in some fixes. I no longer am seeing any
segfaults, and I don't think plobsing has been either. Moritz is still
seeing problems, however.

The problems appear to be GC related, and are therefore triggered by
differences in memory size and layout. We've seen different results
when running with the ms2 GC and the gms GC, and also with/without
--optimize. The problems typically manifest by constant PMCs in
PackFile* constant segments being collected, and causing segfaults
when they are later referenced. These problems seem to only appear
with Rakudo, I haven't seen any reports of similar failures in the
parrot test suite, in NQP, winxed, or elsewhere.

I would really like to get a few more eyes on this problem if we can.
We do have the release coming out next Tuesday. If we can't get this
issue resolved by this weekend, I am going to back out the
imcc_compreg_pmc merge so we can have a clean release. Ideally, I
would like to not have to do that.

Any help, either in reporting new instances of failure or helping to
debug/fix would all be greatly appreciated.

Thanks.

--Andrew Whitworth