Hi,
On Mon, 2026-10-26 at 14:42 +0100, Johannes Schauer Marin Rodrigues wrote:
I have been shown by others that, without an announcement, NEW queue processing
now seems to throw my upload at an LLM. From the review results:
I fully agree with josch here. For all intents and purposes, this is a
blatant waste of maintainer time. When I first saw that NEW review is
now LLM-assisted, I was disappointed, angry, and my already limited
motivation for Debian work further decreased.
As already pointed out, this was not what the GR decided. Having a LLM
taking over a significant portion of NEW review, and auto-rejecting
packages based on that, directly contradicts not only that, but also
externalizes yet more cost on the maintainer who uploaded the package.
I am not gonna reinterate my points from the GR discussion regarding why
LLM usage is wrong,
so I will keep it to why it is wrong in the context of new review.
To quote one example for an LLM output from NEW review:
" [important] [packaging] The whole jxrlib is bundled although Debian ships it
and the previous upload linked it"
While this seemingly makes sense, a LLM can never know if upstream code
changed, so building against
the packaged version could just fail. This is plain wrong and dangerous
advice since the packager knows best, at the end of the day.
Another example:
"
[blocker] [packaging] Build-Depends and Depends name
golang-github-stephens2424-writerset-dev, not found in the archive
"
This is also just wrong. An experimental upload can just (Build)-Depend on
packages not in the archive; it does not matter as long as it
stays in experimental. That is the whole point of experimental.
While humans still make the final decision at the end of the day, I find
it extremly bad taste to hand this over to LLMs. Not just because of the
social, ecological, and societal consequences, but because as shown
above, the reviews are wrong, and an LLM can never be 100% correct.
Furthermore, just introducing this, without any communication *at all*
to the project, leaves a very bad impression.
This directly goes against the collective decision making process, and
feels very intransparent.
Directly dumping a generated generic review text on maintainers feels
highly disrespectful to me.
The main reason I started to contribute, and (later) especially love
Debian was the community. We all are spread around the globe,
and working today for a better operation system.
Or so I thought.
If I wanted the equivalent to an automatic answering machine take over
most of my package reviews I'd never became a DD. I'd just have
contributed elsewhere.
To reiterate from the GR discussion:
We are not here to package everything, neither to churn out code by the
kloc, nor to review packages as fast as possible.
I am aware that the LLM does not reject the package (yet), however the
decision is still massive biased towards the LLM "decision".
PS: Not subscribed to -devel for my own sanity.
best,
--
Matthias Geiger <werdahias>
Debian Developer -- KiCad Contributor
GPG: ECBEDBB607B9B2BE
