Combine versus updating of liveness

[Jan Hubicka]

> On Wed, 6 Mar 2002, Jan Hubicka wrote:
> > Fixing for mainline and branch is probably the attached patch that converts
> > all three beasts to use BB_DIRTY flags for updating instead of maitaining
> > their own bitmaps.  The liveness is updated once after all three passes iterate.
> >
> > Problem is how to fix this porperly for the 3.1 branch?
> > Unfortunately I don't have testcase that reproduce w/o superblock formation...
> 
> I won't claim to understand the problem, but if you're looking
> for test-cases that break in cfg* code, maybe

This is unrelated to the problem.  The patch fixes different
latent bug in combine.
> 
> I've been meaning to look at for a long time :-( but thought
> it'd be fixed with any new cfg* patch, like happened with the
> recent breakage.  (Thanks for fixing that so quickly!)
> I'll probably look at it myself soonish, but perhaps not within
> the next few weeks.
> 
> If I've told you of this before, please forgive.
I will try to do this tomorrow. Java 3.1 and branch failures have pririty
but I am going to do this next today.

Thanks for letting me know!
Honza
> 
> BTW, with your recent patches (LAST_UPDATED: Wed Mar 6 22:17:32
> GMT 2002), I see there are still failures on native and powerpc.
> I get a different set of failures for cris-axis-elf and
> mmix-knuth-mmixware, though.  No need to worry now: I'll get
> back to you if they aren't fixed when the native and powerpc
> failures are fixed.  (Those failures might not even be "yours",
> of course; I'll make sure credit goes where it's due. ;-)
> 
> brgds, H-P