[Mlir-commits] [mlir] [mlir][linalg] Support subtracting accumulation in partial reduction … (PR #214033)

[Renato Golin]

rengolin wrote:

As a meta comment, though, I'd recommend you don't use AI generated PR descriptions and replies to comments. Also be mindful of AI comments in the code, as they tend to be verbose, generally useless and often wrong.

Auto-generated descriptions hint that you may not understand the code you're asking us to review, putting the onus of understanding it on reviewers, and that's not nice. You may use whatever the AI generates as guidance, but write your own description, focusing on what's being fixed and why, not blindly describing the code. Also, make sure to be explicit in the AI usage, as per the policy comments above. I strongly recommend you edit the description now and simplify it to just what we need to know (the PR is reasonably simple in scope).

In the same way, an AI generated reply to a review comment feels impersonal. Imagine spending time looking at a piece of code and some AI generated description, then adding review comments, only to be faced with more AI replies. Makes people really question looking at the PR again. If you want to participate in the community, spend time writing your replies. Even broken English is better than AI English.

Finally, whenever you get auto-generated code, invariably the model will insert a lot of comments. They're almost always redundant, often useless, occasionally broken. Always double check if the comments relate to the actual change you have, and validate that it will not mislead future developers into what the code is actually doing.

https://github.com/llvm/llvm-project/pull/214033