Kyle Cripps

> The way to test this is to remove the passes that were introduced to tree-ify the DAG before def-use (e.g., Cloner in simplifyDefUse.h). Unfortunately not all of them can...

> > The way to test this is to remove the passes that were introduced to tree-ify the DAG before def-use (e.g., Cloner in simplifyDefUse.h). Unfortunately not all of them...

> If I understand what you are saying, this means that the context representation you are using is incomplete. You need to keep the child number as well. It may...

> > It may be incomplete, but it is still an improvement over the current implementation, so additional improvements can be made separately. And the future improvements will need to...

This PR should be ready now. Please see https://github.com/p4lang/p4c/pull/4810 for the implementation of "some flavour of value numbering" suggested by @asl.

> @kfcripps Are there updated benchmarking results? Or the ones above are still valid and there were only correctness changes? @asl New results on this branch (as of 50c0f9fbe55e969c167afac3d8204b34348163bd): test/gtestp4c...

> What about #4385? @fruffy Although the problem described in #4385 was introduced by the same offending PR as the other mentioned issues, it does not necessarily mean that the...

> A test case for #4507 seems to be missing here. See https://github.com/p4lang/p4c/pull/4797#issue-2406024872.. Mihai's PRs masked the root problem, so I do not know of any test cases that pass...

See https://github.com/p4lang/p4c/issues/4811

@mihaibudiu By API do you mean `visit_unique_children()`? I thought it might be useful for other passes that also do not want to traverse equal child nodes, but if you insist,...