[tree-ssa] bootstrap failure

[law@redhat.com]

In message <3EA5D12C-2070-11D8-908F-000393A6D2F2@physics.uc.edu>, Andrew Pinski
 writes:
 >
 >On Nov 26, 2003, at 16:15, law@redhat.com wrote:
 >
 >> In message <1069870197.3158.6.camel@frodo.toronto.redhat.com>, Diego 
 >> Novillo writes:
 >>> There's definitely more than one bad interaction going on between PRE
 >>> and the new PHI manager.  After fixing the allocation problem, I'm 
 >>> still
 >>> getting various SEGVs on alpha, ia64 and amd64.  They all happen
 >>> somewhere in the GC, as it tries to mark roots.  It doesn't happen on
 >>> any of the 32bit hosts, though.
 >
 >Actually I think it can be reproduce on a 32bit host, see PR 13193,
What does the backtrace look like?  There's two known issues right now.  The
first is that PRE can request a PHI node of length one.  That can cause a
crash in tree-phinodes.c.  We know where and how to bullet proof tree-phinodes
from this sillyness.


The other is a crash within the collector itself, which is the "interesting"
problem.

If you crash is in the collector, can you send a backtrace my way privately.
That would be more data than I've managed to extract from my alpha so far :(

 
 >would try i386-pc-linux-gnu.
Running flawlessly.


 > like this, most likely it just something is not initialized to 0 aka 
 > not cleared.
Most likely, yes.  Though the code is good about wiping the lists clean.

Jeff