IRC channel logs

2024-05-03.log

back to list of logs

<PurpleSym>Standing knee-deep in dependencies of dependencies myself I wonder what we can do about that. Perhaps the Docker faction is right and traditional distributions are dead, because they simply cannot (with reasonable effort) provide a coherent set of packages any more.
<PurpleSym>Maybe we have to embrace that approach and reduce Guix to a minimal set of packages needed to build Guix itself and then build (possibly incompatible) channels on top of that.
<civodul>with all the dependency metadata now available out there (wheel, requirements.txt, poetry whatever), why aren’t we able to get a consistent set of packages?
<PurpleSym>It would lead to alot of stupid and unnecessary duplication, but throwing everything away and running `guix import` from scratch (“the Docker approach”) seems easier than carefully breaking and unbreaking dependencies 10 levels down the stack.
<civodul>it works for crates and for CRAN
<PurpleSym>civodul: Because that metadata is as bad as the software.
<civodul>ah :-)
<civodul>but then Python is doomed?
<PurpleSym>Kindof.
<PurpleSym>Not Python itself, but the ecosystem.
<civodul>do people use pip exclusively with wheels, pre-compiled, and thus they don’t realize dependency info is all broken?
<PurpleSym>Yep. Even on commercial projects we just `pip install` the wheels into our Docker images (without checking the code, obviously).
<PurpleSym>I think Guix is one of the few distribution that even tries to run tests. We could save us alot of trouble if we did not.
<civodul>what a bleak situation
<PurpleSym>Not bleak, just a reality of finite resources that need to be prioritized (time, money, knowledge).
<PurpleSym>Just like I exhausted my Guix time for today :(
<PurpleSym>rekado: Can we drop the python-notebook updates from python-team and just merge the improvements to pyproject-build-system? I don’t have time to go down that rabbit hole and if no-one else does, the branch will just bit-rot even further.
<civodul>PurpleSym: oh sure; i was thinking about the situation for Python in general, it’s bad that with the profusion of packaging/environment tools they have the situation has not improved
<civodul>there’s no lack of resources in Python, as i see it
<civodul>i understand why you’re exhausted though!
<PurpleSym>civodul: With “profusion of packaging tools” you mean PEP 517, i.e. pyproject-build-system?
<PurpleSym>Not exhausted. A few hours a week are simply not enough to deal with this.
<PurpleSym>We need more full-time Guixers :)
<civodul>heh
<civodul>PurpleSym: i mean requirements.txt, wheels metadata, Poetry, etc.
<civodul>i don’t know the PEPs :-)
<civodul>seeing it as an outsider
<civodul>but yeah, this work certainly needs more people dedicated every day