Allow more granular control on which artifacts are output

[NekkoDroid]

This PR repurposes SplitArtifacts= to be able to specify multiple artifact types that should be split out of the image. By specifying one or more of uki, kernel, initrd and partitions the respective components are extracted from the image. The truthy and falsy values still map to the equivalent of the old values to preserve backwards compatibility.

[behrmann]

I like the refactor in the first commit (sans the style nits), but I'm a bit -0 on the second commit, since it breaks backwards compatibility and I'm not quite sure I understand why the superfluous artifacts cannot be deleted with a postoutput script, but I don't have too strong feelings there.

The backwards compat could probably be kept by making ArtifactsOutput a enum.Flag instead of a StrEnum, which would then probably look something like

class ArtifactOutput(enum.Flag):
    false = 0
    uki = 1
    kernel = 2
    initrd = 4
    partitions = 8
    true = 7

[NekkoDroid]

still looking into using enum.Flag, seems promising but I am still having a few problem with it :)

[NekkoDroid]

I am unable to figure out how to serialize the value to JSON and actually make it assign the value when returning from the parse function. When setting any value it still keeps resulting in the default (ArtifactOutput.no) when printing the summary

[NekkoDroid]

the JSON problem is "solved" (happy for any recommendation that makes it easier to understand than the raw flag number).

The issue with not being able to set the values in the config still persists though. (I was looking at the wrong config 😓)

[NekkoDroid]

Some CI fails due to dnf, but others seem to not be able to in an IntFlag :/ Unsure what is causing this

  Traceback (most recent call last):
    File "/home/runner/work/mkosi/mkosi/mkosi/run.py", line 64, in uncaught_exception_handler
      yield
    File "/home/runner/work/mkosi/mkosi/mkosi/run.py", line 105, in fork_and_wait
      target(*args, **kwargs)
    File "/home/runner/work/mkosi/mkosi/mkosi/__init__.py", line 4416, in run_build
      build_image(
    File "/home/runner/work/mkosi/mkosi/mkosi/__init__.py", line 3661, in build_image
      if ArtifactOutput.partitions in context.config.split_artifacts:
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  TypeError: argument of type 'int' is not iterable

[behrmann]

I'm travelling this weekend and forgot my notebook, so will only be able to look at this again on Monday, sorry.

[NekkoDroid]

Needing to use ArtifactOutput(split_artifacts) to get CI to be happy isn't the prettiest but I guess it gets the job done. If anyone knows what is causing it or how I could fix it other than what I did I'd like to hear.

[DaanDeMeyer]

Can't we just return the path here and then get rid of get_uki() and use copy_uki() everywhere instead? I'd prefer to have this be a single function instead of two.

[NekkoDroid]

That would result in the uki always getting copied when it exists, even when it wasn't specified. Just conditioning the copy part and still return the path would feel a bit weird, but I could do that.

[behrmann]

I think it's good as is with the name change. I don't think copy_uki shouldn't copy the uki in some scenarios, that would just be false advertising. I could see just always copying out the UKI and having the kernel and initrd be options in addition to the uki. I think that would also be fair, but I lean more towards the way it is now.