zivid / zivid-python

Official Python package for Zivid 3D cameras
BSD 3-Clause "New" or "Revised" License
40 stars 14 forks source link

Fix broken CI #214

Closed knzivid closed 1 year ago

knzivid commented 1 year ago

Archlinux moving to gcc 13 triggered a series of cascading issues

knzivid commented 1 year ago

You are right, the source distribution is broken on arch. This PR makes the CI green and does not fix the customer situation. I have a follow-up in mind to precisely control the conan user home and override it from setup.py but my first goal is to make zivid-python green. If possible, I would remove conan completely from zivid-python because it is a moving target that would break again.

When a customer installs zivid-python on a new enough distro that uses gcc 13 or if clang releases a new version that's not tracked in conan's settings.yml, zivid-python will break. This is a pitfall on all platforms.

conan may also decide to drop support for v1, in which case we need to either bump to v2 or maintain our fork of settings.yml with new compiler versions. Using conan is not worth it.

As the CI triage person for this sprint, I can only take detours into fixing red pipelines if they are simple enough. If you don't accept this PR, I will have to close it and I can't guarantee I will get back to the perfect solution soon.

nedrebo commented 1 year ago

I IRL'ed with @eskaur.

The problem here, as far as I understand it, is that Arch moved more quicker than Conan. And this is probably a transient error. However, one that appears now and then in various forms. It can also be Arch is too new for some pip packages now and then.

We agreed that Arch testing is not the important thing here, and it is not worth the random breakages.

What is important is that we test on a variety of OSes, and not only Ubuntus to make sure we do not have Ubuntu-specific code.

Instead of this PR, @eskaur will do the following within a day or two. Before that it is ok that SDK team ignores CI errors from this repository.

knzivid commented 1 year ago

I agree with the other points but not with the claim that this is transient. This is systematic and will break if a customer uses a different compiler as the system default irrespective of the distro.

Happy to discuss it irl