This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [tree-ssa] COND_EXPR lowering.
- From: Daniel Berlin <dberlin at dberlin dot org>
- To: law at redhat dot com
- Cc: Zdenek Dvorak <rakdver at atrey dot karlin dot mff dot cuni dot cz>,gcc-patches <gcc-patches at gcc dot gnu dot org>,Andrew MacLeod <amacleod at redhat dot com>,Diego Novillo <dnovillo at redhat dot com>
- Date: Mon, 27 Oct 2003 16:49:26 -0500
- Subject: Re: [tree-ssa] COND_EXPR lowering.
- References: <200310272140.h9RLegDj009018@speedy.slc.redhat.com>
Err, ummm, I thought Daniel fixed this in another way last week!
No, unfortunately, this is a different bug, and would occur on the
mainline if we ggc_collect after recalcing dominators, and still
expect the dominator info to be valid after ggc_collect.
IE you can trigger this problem (on the mainline) by doing:
1. You calculate dominance info
2. You call ggc_collect
3. You try to access the same dominance info
Apparently, dominator info is ggc_alloc'd, but not marked from anywhere
at all.
This is bad.
Last week, I fixed dom_children being ggc_alloc'd but not marked from
anywhere at all.
I certainly wouldn't consider this an "obvious" patch -- just look
at the discussion which led up to it -- anything with that long of
a discussion certainly doesn't qualify as obvious to me.
He effectively is just not ggc_alloc'ing the dominance info array
anymore, which is one way to solve the problem. Another way is to have
it marked from somewhere, but i'm not sure where we'd do that from.
When you return, please look at what Daniel did and either remove your
patch if it has become redundant or explain why your patch is still
necessary.
jeff