An old kernel library had four bugs. An LLM found three of them.

Alongside Linux 7.3-rc7, Linus Torvalds published an assoc-array-fixes tag. Nils Ollrogge had reported a bug in the kernel’s associative-array library and sent a patch. Torvalds said the code was too obscure to follow by eye, so he used current LLM tools to make the change readable, which is the approach he has been telling other developers to try.

The assisted branch splits some of the larger functions so a person can see what they do. That rewrite was then run through a user-space test harness. The harness confirmed the bug Ollrogge reported. It also found three more, all present since the library was added in 2013.

The associative array is not a toy structure. The kernel uses it in real paths, and a bug that survives for thirteen years is usually one that the existing tests never reached. Making the functions smaller did not fix the logic by itself. It made the logic testable. The extra bugs showed up because the harness could exercise the library without booting a kernel.

Torvalds has framed this as a reading aid, not as a reason to merge unreviewed model output. The branch is aimed at the Linux 7.4 merge window, which means it still has to survive the usual list review. The interesting part for other maintainers is the sequence: a human bug report, a model used to unpack code nobody wanted to reread, and a harness that checked both the known bug and the ones nobody had filed.

If you maintain a similarly dense subsystem, the transferable piece is the harness. The model made the patch discussable. The tests are what found the decade-old mistakes.