Gard Spreemann dijo [Thu, Aug 13, 2026 at 12:17:29PM +0200]:
What do you think about other software-written packaging? i.e. about
packages created by our usual dh-make-* tools? I have used them quite
a bit. Of course, I do take a look (and fill in with more data where
needed). I have even written one dh-make-*¹ tool several years ago; I
didn't use it much for Debian work IIRC, but had quite a large
deployment of debianized Drupal modules at a private apt repo.
¹ https://manpages.debian.org/testing/dh-make-drupal/dh-make-drupal.1.en.html
Yes, it would be excessive to label dh-make-* as “rule-based expert
systems”, but to an extent... yes, it's a software-based shortcut that
saved me a lot of time filling in very repeatable packaging patterns.
I don't see these as anywhere near causing the same asymmetry in effort
as AI tools.
Maybe I'm being too simplistic in my thinking.
Maybe I'm being simplistic as well, but let me explain myself better
here.
If I am contributing something to Debian, it must be something I feel
to be _able_ to understand and feel responsible about. That is, if I
do a NMU, I must be able to understand (and defend!) its diff. If I
introduce a new package, I must understand the debianization delta
against upstream.
If I ask an LLM to package something for me, I should take a good,
critical dive into whatever it is proposing; lets say at the very
least with as much detail as a package I'm being asked to sponsor (or
more than that, as I cannot say the LLM has the “intent” of behaving
well and doing things right and up to the latest standards).
If I were to apply this guidelines to my work, would you accept my
LLM-generated, human-reviewed contributions?
