[mlir] Fix nested codeowners for vector dialect

[Groverkss]

Due to how CODEOWNERS works, if a file matches two rules, the later rule will take precedence.

Because of this, /mlir/include/mlir/Dialect/Vector rules were not matching against /mlir/include/mlir/Dialect/Vector/IR .

This patch adds users in the directory rules, to the finer grained rules, which I'm guessing was the intended effect that they expected when they were added.

[kuhar]

Gah, this sounds like quite a footgun...

[banach-space]

Thanks for the fix!

1. Do you have a link documenting how CODEOWNERS works? It would be good to add that in the summary.
2. Shall we keep the list of reviewers in alphabetical order?

[dcaballe]

Yeah, I also went through this issue a few times. My opinion is that we should limit the rules to dialects or well-defined coarser-grained components. Having rules per .h/.cpp file or small subfolders is too fine grained and introduces too much noise in the CODEOWNERS file.

[Groverkss]

Yeah, I also went through this issue a few times. My opinion is that we should limit the rules to dialects or well-defined coarser-grained components. Having rules per .h/.cpp file or small subfolders is too fine grained and introduces too much noise in the CODEOWNERS file.

I think that's a good idea. There is a bit of free for all going in the file, I don't think there is anything governing how rules are added there.

[Groverkss]

1. Do you have a link documenting how CODEOWNERS works? It would be good to add that in the summary.

Yes, it's in github docs: #118208 I can raise a new pr to add that into the CODEOWNERS summary

2. Shall we keep the list of reviewers in alphabetical order?

Sure.

[banach-space]

Yes, it's in github docs: #118208 I can raise a new pr to add that into the CODEOWNERS summary

Wrong link :)

My opinion is that we should limit the rules to dialects or well-defined coarser-grained components. Having rules per .h/.cpp file or small subfolders is too fine grained and introduces too much noise in the CODEOWNERS file.

+1 I propose having one rule for Vector. For finer-grain control, we could use GitHub sub-scriptions.