https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79405
--- Comment #11 from Aldy Hernandez <aldyh at gcc dot gnu.org> --- I had a brief interchange with Segher to see if he was still working on this (no). He had this to say (paraphrased with permission): [There are two proposed ways to make this better: 1) Propagate things in an order that does not miss opportunities and always terminates. This requires big code changes. This would try *all* possible propagations until a fixed point is reached (the current code does not try *all* possible propagations). 2) Construct a DAG (or a linearisation even) and only propagate according to that. Also a big code change, but a big simplification. It does however lose opportunities. This is Richi's suggestion in comment 9, which may miss propagations (it cannot propagate around loops). This may not matter much though. Neither is suitable for stage 4 I'd say.] I agree that these do not look like solutions that are stage4 material. Furthermore, I don't see the current stop gap as a problem, given that it only affects a testcase with undefined behavior. Given this, I don't see how this bug can be classified as a P1. Thoughts?
