Hi,
On Tue, 27 Oct 2026 at 23:23, Ludovic Courtès <[email protected]> wrote:
> Ricardo Wurmus <[email protected]> skribis:
>
>> Vibed software that Ludo mentioned is like abandoned pet
>> projects. This doesn't really have a place in Guix. Most of these
>> pieces of software are disposable. We have rejected other software in
>> the past out of concerns for a lack of quality and poor upstream
>> practises.
>>
>> This is not primarily about the fact that the authors (human
>> triggers?) use LLMs; it is just that this is what makes the software
>> so disposable as to be unfit for Guix.
>
> đź’Ż
>
> “Poor quality” can be a criterion but it’s more subjective than
> “vibe-coded”; lack of popularity is not necessarily a good criterion,
> except in the extreme case of single-user vibe-coded software.
>
> I think, as a first step, what we could do is agree that there are
> packages that we, committers, may choose to reject on the basis of
> criteria such as these.
The package [0] had been included in Guix on Aug. 2025 and might be
LLM-assisted since the very beginning. Then, the package had been
updated 3 times; probably again LLM-assisted.
Somehow, it passed a subjective criteria 4 times and now this subjective
criteria has changed so it’s rejected. It’s confusing, no?
The volume of LLM-assistance might have changed of scale, though.
Well, I’m totally fine that we, committers (or Team members?), may
choose to reject contributions. However, what I’m asking for is to
sketch a shared understanding about the criteria (singular or plurals).
Look, we write the term quote vibe-coded quote where I’m not even sure
we put the same vague meaning behind the term. :-) I mean, the
acceptance or rejection cannot depend too much on the subjective lens of
one or the other committer.
Moreover, please note the variant of the question [1,2] discussed some
weeks ago. The proposition had been to add the label (tag)
“kind/gen-ai-suspect“. Maybe it could just be “package-ai-suspect”?
All in all, concretely we could agree on a rule like:
1. A reviewer marks as “package-ai-suspect“ a package contribution that
looks suspicious.
2. The reviewer pings another committer for an opinion.
3. If the suspicion is confirmed by 2 people, then the package
contribution is rejected with a pre-formatted message.
4. If no one provides a second opinion, then after 3 months the
package contribution is closed.
WDYT?
Rejecting without asking for another opinion appears to me harsh because
mistake happens.
We could add a warning note in the manual section “(guix) Reviewing the
Work of Others” [3] detailing this rule; maybe also containing the
pre-formatted message.
As a janitor of the venerable guix-patches tracker, a rule of thumb
based on my observations seems: if a contribution about a package isn’t
done after 1-4 months, then this contribution will never cross the final
line; it’ll be one the almost 1k still open patches (without counting
the 1.1k still open PRs).
Cheers,
simon
0: https://codeberg.org/guix/guix/pulls/12002
1: GenAI tag on Codeberg
Sharlatan Hellseher <[email protected]>
Sun, 11 Oct 2026 23:06:39 +0100
notmuch:id:[email protected]
https://lists.gnu.org/archive/html/guix-devel/2026-10
https://yhetil.org/guix/[email protected]
2: Re: Re: GenAI tag on Codeberg
Sharlatan Hellseher <[email protected]>
Mon, 12 Oct 2026 20:01:55 +0100
notmuch:id:[email protected]
https://lists.gnu.org/archive/html/guix-devel/2026-10
https://yhetil.org/guix/[email protected]
3: https://guix.gnu.org/manual/devel/en/guix.html#Reviewing-the-Work-of-Others
PS: BTW, I’m more than convinced that most of our discussions about
genAI and other topics would be drastically simplified if we had
sketched somehow our project “ethical values” (principles) about
what means “Free Software” for our community. Well, I’ll not raise
again this point on the Guix channels of communication since the
(frictional) bottom-up path seems preferred over a (complicated)
top-bottom path. Both are equally good or bad; only a
preference – I can live with. :-)
