Re: [PATCH net-next 2/2] docs: netdev: clarify expected interactions with LLMs

[Paolo Abeni <pabeni@redhat.com>]

On 7/24/26 3:39 PM, Matthieu Baerts wrote:
> Hi Paolo, Jakub,
> 
> On 24/07/2026 13:56, Jakub Raczynski wrote:
>> On Fri, Jul 24, 2026 at 11:24:05AM +0200, Paolo Abeni wrote:
>>> +both use the Sashiko infrastructure on top of different models. Reviews are
>>> +available after 24h. Patch authors are expected to proactively look into the
>>> +AI-generated reviews and handle such feedback as any other kind of review:
>>> +either debate it or address it. In both cases a reply on the mailing list is
>>> +expected.
>>> +
>>
>> Would it be reasonable to implement automatic response from Sashiko to
>> mailing list? Some mailing lists (eg. dri-devel as far as I have seen) post
>> all replies from Sashiko to thread, which is imo not bad,
>> although it could be filtered by High/Critical severity.
>>
>> What do you think about potential for similar solution on netdev,
>> since AI review has become 'basic' check in patchwork?
> I was thinking that too: it would be easier to reply to the comments,
> and help the authors to remember about them. That will not help to
> decrease the traffic on the netdev ML, but it is also possible to
> configure Sashiko to email only the authors:
> 
>   https://github.com/sashiko-dev/sashiko/blob/main/docs/configuration.md#email_policytoml

One concern I personally have with the above is that it could turn the
ML in an public automated AI review tool, definitely flooding any
eventual good contents.

In any case this discussion is IMHO quite orthogonal to the
documentation update.
>>> +Authors are strongly encouraged to run LLM reviews on the posted patches in
>>> +advance of the actual post. Large series triggering a significant amount of
>>> +AI-generated feedback will likely get little attention from maintainers and
>>> +reviewers.
> 
> If Sashiko emails the list and/or the authors, it would be good if this
> note is added to the emails, not to push people to abuse the public CI
> and AI resources.
> 
> On a similar topic, it might be interesting to mention SUIE and the
> scoring system. Or mentioning that sending multiple untested versions
> will move their series to the bottom of the queue.
I have some doubts WRT setting in stone (in the doc) the actual patch
review order or an explicit "sanctions" system. I think the patch
wording allows for more flexibility.

/P