Reduce reliance on main "canonicalize" pattern
- Dominant language
- MLIR
- Stars
- 906
- Forks
- 171
- Avg merge
- 4d 12h
- Merged PRs (30d)
- 32
Description

I think the correct thing here is to do the following:
- where we rely on specific canonicalization patterns as an integral part of the pass pipeline, put those patterns in a specific name passed, so that if they ever get removed/etc upstream, we don't get sudden failures without warning)
- Verify that all pipelines still produce code succesfully if the canonicalization is removed/replaced with no-op
- how to efficiently set all canonicalize passes to no-op? (Maybe overwrite the definition of addCanonicalizerPass to be no-op?)
- how to efficiently check that the output is still "correct" even though it's now syntactically different (i.e., FileCheck tests will fail?)
Contributor guide
Assessment
This issue has not been assessed yet.