One nasty thing I've caught reviewing LLM-authored changes (in a Rust codebase at work) is the LLM adding #[ignore] to existing tests (and obfuscating it if you tell it not to do that, via things like #[cfg_attr], comment markers or if false), or even deleting them outright, while adding a decent number of new tests that make sense.
If you don't spot what it's done to your test suite, you get surprised because the patched code compiles and passes all tests, and yet some of the things you "know" you test don't work.
I'd expect this to be even worse for the kernel, where tests are often wrong if they exist at all, and so an LLM producing a plausible comment explaining why the test is bad will result in someone accepting the removal at face value.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。