tqchen commented on PR #13916: URL: https://github.com/apache/tvm/pull/13916#issuecomment-1419367701
Thank you, everyone, for chiming in. In order to get unity merged into main, we will need to ensure the end state (at the merging proposal) of the unity branch gets to a ready state. In the meantime, the branch development offers some flexibility for us to explore ways to gain agility with quality. We can have continuous conversations through the course of branch development to iterate the modules so we can answer the questions and continue to iterate on the foundation before merging the proposal. There is no disagreement on goals, such as having high-quality solid foundations while collecting community inputs not only at the PR merging point but through continuous periods of time during development. For the branch establishment period specifically, some initial agility is helpful to get the initial scaffolds and enable quicker migration can benefit the community. We can also leverage the branch development opportunity to have continuous conversations instead of stopping at PR merges. For example, suggestions about changes, questions in this thread, or follow-up thread (aka, nothing is yet final at the merging point, and the community can continue to provide inputs and iterate on the solutions). To make sure we have accountability for quality and continuous community participation, the merger should also be responsible for addressing these follow-up inputs, suggestions, and questions and resolving foundational technical debts that the merger may have caused. This is a bit like a mixture of review-commit-review models. We can also continuously monitor the experiments and see, for example, if there are any lessons like technical debts that do arise, and adjust accordingly. Coming specifically to the technical part of the current PR. Please do bring followup suggestions at any time point, in this thread or other places about the particular change. The initial suggestions about test coverage is very valuable and I see yuchen have updated the proposal set with proper test cases. Suggestions of possible additional test cases or possible technical debts are very welcomed. At the moment the test covers the struct info analysis and expr construction, hopefully more tests will be enabled after the parser is enabled through a rich set of TVMScript. Thank you -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
