On Mon, 20 Jul 2026 at 08:22, Peter Maydell <[email protected]> wrote:
> Reading this comment prompts me to check with you whether you have
> read our AI contribution policy, which at present is essentially
> a "no AI generated code or text in contributions" policy:
> https://www.qemu.org/docs/master/devel/code-provenance.html#use-of-ai-generated-content

Thanks Peter, and fair question -- the answer is yes, AI tooling was
used in producing this series, so under the current policy I should not
have sent it in this form. I apologise for that, and I appreciate you
flagging it directly rather than just dropping the patches.

To be transparent about what that means for these patches: this work
comes out of a project that runs macOS guests on QEMU/KVM, and I use an
AI coding assistant in that project. The comment you quoted is a good
example -- it's essentially the assistant's working notes shipped
verbatim, which is exactly the kind of thing that shouldn't be in an
upstream submission (and, as you say, the development-history narrative
doesn't belong there regardless of how it was written).

What is *not* AI-derived is the hardware ground truth: the USB
descriptors, HID report descriptors and endpoint topology are captured
byte-for-byte from real Apple peripherals, and the design decisions
(which device to emulate, why Mighty Mouse rather than Magic Trackpad,
how it binds in macOS) were mine. But I recognise that doesn't resolve
the DCO clause (b) concern about the generated code and text itself, and
I don't want to hand-wave that.

So I'll withdraw this series for now rather than ask you to review it
under a cloud. If I reimplement the device code and comments myself,
from the captured hardware descriptors as the spec, would a resubmission
be welcome -- or would you rather this category of contribution wait
until the AI policy itself is revisited?

Not looking to litigate the policy -- just want to be straight about
provenance and do the right thing here. Thanks again.

-- Matthew

Reply via email to