Scanning the linux-kernel mailing list just now (a bad habit I have), I see patches to add significant features to the Alpha architecture, convert Perl scripts that haven't been touched in decades to Python, and replace user and group IDs with UUIDs.

Not sure how we ever got by without these wonderful new tools.

@corbet no patches to relicense to CDDL? 🙃

@corbet Alpha needs to die and let's definitely not add any new features to museum architectures.

@corbet um, the implication that Matt Turner's work is slop is surely undeserved

@corbet I wonder how often the Python rewrites will need to be touched going forward for newer versions of Python, and of course then won't work on older versions. #sigh

@corbet seriously disturbs the natural rule that stuff that nobody cares gets removed because it's marked BROKEN for a while. Will end up maintaining stuff for longer than it's needed because LLMs will keep patching without actual users

Something like wonder, anyhow.

@corbet

@amonakov @corbet I have indeed used Claude Code for Alpha kernel work (and the patches are tagged with `Assisted-by: Claude:claude-opus-5`, etc).

The work going into v7.3 from me is

(1) some bug fixes to math-emu found by running the glibc test suite.
(2) some locking fixes found by enabling lockdep on alpha (by Magnus Lindholm)

Pretty small potatoes, IMO.

I have used Claude, especially for debugging, but I would hope no one would call the results slop.

@mattst88 @amonakov Stuff on the lists, though, includes self-extracting compressed kernel images, perf improvements, and 700 lines of Tsunami machine-check work. The potatoes are growing :)

One can argue that this work is low-risk and hurts nobody, but one can also certainly ask whether the mainline kernel should still be adding features to a long-dead architecture.

@corbet @amonakov Oh, yes. We've certainly got a lot larger contributions for v7.4. But I think they're either done without LLMs (GENERIC_ENTRY and the Tsunami machine check work -- though it's had LLM review and probably an LLM was used to split the initial patch into a reviewable series), or is reducing divergence between Alpha and other arches (kernel compression, trivial Kconfig options, perf tool improvements)

Dunno. I guess you can make the case we should just delete it all instead. :(