An ext4 driver arrived at OpenBSD with no author↗ (external link)

Jonathan Corbet on a submission to OpenBSD's tech list: Thomas de Grivel sent in an ext4 implementation with full read and write support that passes e2fsck, lacking only journalling. He was straightforward about where it came from — generated with ChatGPT and Claude, with, in his words, careful code review and error checking on his part.

It was rejected, and the reasons are more interesting than the rejection.

Theo de Raadt went to provenance: "the chances of us accepting such new code with such a suspicious Copyright situation is zero." Damien Miller pointed at the liability of shipping code whose ownership is unsettled. Christian Schulte raised the GPL-licensed ext4 documentation and what that might mean even for an independently written implementation.

But the objection that will outlive the legal ones came from Tyler Anderson, who asked who maintains code when its original "author" does not understand it and has no personal investment in it.

That is the question, and it is not really about AI. A filesystem driver is a thirty-year commitment. Somebody has to be there at 2am in 2034 when a corruption report comes in against a code path nobody has looked at since it was written. The reason projects care who wrote something is not credit — it is that authorship has historically been a decent proxy for somebody understands this and will still be here. Generated code severs that proxy in a way a copyright-assignment form cannot restore.

Worth noting what this is not. It is not a claim that the code is bad. It might be excellent; passing e2fsck with read-write support is not nothing. OpenBSD declined it anyway, because a project that accepts code on the strength of "it seems to work" has already lost the thing that makes it OpenBSD.

The copyright argument will get resolved eventually, one way or another. The maintenance argument will not.