Usually we check for error_mark_node first; with that change the patch is OK. I note that Diego already checked in a similar patch on the tree-ssa branch. I don't know why he didn't check it in on the trunk as well. Jason