Alex Crichton (Oct 19 2026 at 20:43): Alex Crichton (Oct 19 2026 at 20:43):As LLM-generated changes continue to rise I wanted to drop a reminder here both for us as reviewers and for others -- don't forget to review the tests that are generated. I know I often skim quickly over them myself and they sometimes take awhile to go through too. That being said, we're going to quickly run out of CI resources if every change, regardless of what it does, comes with 10x more lines of tests than code changes.
In my experience LLMs still need to be coaxed to produce tests that I subjectively find landable. For example they'll often generate tests that are extremely precise to whatever problem is at-hand. I find such tests not very useful and often delete many of them.
Of course don't stop making tests, tests are still good, but I'm mostly reminding authors/reviewers to be judicious and double-check that tests are necessary. If they're necessary, fantastic let's land them. If it's instead testing 10 permutations of what is basically the same thing, maybe we only need one of them -- things like that.
Also where possible *.wast and *.clif tests are the best. Those take no extra time to compile and only take a little time to run, and we can easily shard those and figure out runtime. *.rs tests are the biggest since we pay primarily for compile time.
Last updated: Oct 31 2026 at 23:03 UTC