[Mlir-commits] [mlir] [mlir][python] Fixup the constantTypes of pdl.TypesOp (PR #75812)

[Oleksandr Alex Zinenko]

ftynse wrote:

> but it feels like this is a bug in how pdl_interp.check_types handles []. @ftynse do you know what pdl_interp should be doing here?

Nope.

> Thanks for your reply. I get this message from -debug information but not understand the meaning well.

This message has absolutely nothing to with Python or PDL. It's debug information from the transform dialect, as indicated by the prefix. It is irrelevant unless you are suspecting an error in the transform dialect.

> I'm not sure if it's a bug of pdl_interp.check_types or pdl.py as well.
> If we replace the %results = types with %results = types : [], the transformation fails.

Since this can be reproduced with `mlir-opt`, this has absolutely nothing to do with Python bindings. Any fix to bindings will not in anyway resolve the issue.

Also, can you elaborate what kind of issue you are experiencing? I have no problem running either original or modified version of this test at llvmorg-18-init-15136-gebd0b6bbb462. The modified version doesn't apply tiling because the op fails to match. This is expected since you ask it to match an explicitly empty list of results (`[]`) and the only matmul op has one result.

https://github.com/llvm/llvm-project/pull/75812