Hzfengsy edited a comment on pull request #7630: URL: https://github.com/apache/tvm/pull/7630#issuecomment-803371685
Thanks everyone's kindly input. It makes this PR much clearer and readable. I totally agree that great documentation is necessary for open-source projects. But I agree more with @junrushao1994 that we should distinguish who the documents are written for. We cannot write every doc for new users without any background. I insist that the docs in TVM parser are written for developers who have known a lot with tir and TensorIR. So that documents for data structures (e.g. block signature: reads/writes/match_buffer_region) are not necessary. But I have been making my effort to address actional items as much as possible, making things move on. Finally, I add the block example in `context_maintainer` to make everyone happy, although I still disagree it should be here. I have polished the code for many rounds together with my co-authors. I want to thank their efforts again for improving code quality. During development, I found some latency issues of TVMScript Parser. I just updated the following thing besides basic functional improvement: 1. Type annotations for the Parser 2. Fix error reporting issues left by last PR (anothered by @tkonolige) 3. Add many test cases for Parser and error reporting (not only for new features of TensorIR) 4. Add more document and assert checks lacking in the current version This PR's original purpose is to add support for TensorIR, which is long about 1000 lines of code. But I just tried my best to improve code quality, and finally, the changes became to 2000+ lines. That's far more than what should an incrementally PR does. I'd love to try my best to polish code but not endless requests. In general, I'm happy to improve code quality and documentation. But it still to think about the tradeoff between readability and heavy documents. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
