On Tue, Dec 13, 2011 at 1:00 PM, Jakub Jelinek <ja...@redhat.com> wrote: > On Tue, Dec 13, 2011 at 11:52:38AM +0100, Richard Guenther wrote: >> On Mon, Dec 12, 2011 at 6:01 PM, Jakub Jelinek <ja...@redhat.com> wrote: >> > In gimple_fold_call (called from fold_stmt from replace_uses_by from cfg >> > cleanup) we weren't calling maybe_clean_or_replace_eh_stmt, so when >> > we've replaced a printf call (which can throw) with puts (which can throw >> > too), nothing would update EH stmts. It would be problematic calling >> > gimple_purge_dead_eh_edges, because callers might be surprised by that, >> > especially when this happens during cfg cleanup, so instead I just assert >> > it is not needed and don't try to fold if a throwing stmt would be replaced >> > by non-throwing. FAB pass can handle that instead. No folding >> > has been actually disabled because of that check during bootstrap/regtest, >> > so it is there just in case. > > Note, I've already committed the patch yesterday.
Yes, I've seen that. I think it's ugly and inconsistent - we can certaily fold non-call stmts from throwing to non-throwing. Think of -fnon-call-exceptions and x + 1.3 and folding after propagating 1.0 to x. It is not fold_stmts business to care about EH info or edges (it works on statements in isolation). >> I think that all callers of fold_stmt are supposed to handle EH >> updating themselves, similar to how they are supposed to >> call update_stmt. I see that replace_uses_by does call > > Some do something, but others don't. Do you think it is preferable when > the callers do that (all of them)? Even if that is chosen, while some could > purge dead eh edges, other places can't do that IMHO, so either we need to > simply not do transformations that remove EH edges in fold_stmt, or have an > argument that is passed down whether it is ok to do so. I think all callers can (should be able to) do that. In the light of -fnon-call-exceptions we probably miss quite a few - but we should fix them. [Unfortunately some of our propagation helpers do not, or do not have a good way of returning whether EH/CFG cleanup is necessary, notably gsi_remove, which removes the stmt from the EH tables but does not purge edges.] Richard.
