Add executable mkosi.version support for generating the version dynamically

[NekkoDroid]

The script is executed unsandboxed (not a fan of this, but unsure how this should be done before any configuration even exists) and reads stdout to figure out the version.

First commit allows more configurable handling of "standard config paths". Maybe this change should also be extended to mkosi.rootpw to allow generating it dynamically?

Closes #2915

[behrmann]

Having both mkosi.version and mkosi.version.run leaves the path open to reusing the former as input for the latter, as I described earlier, but if we decide we don't need that, this could be conveniently folded into one and use mkosi.version as a script if it's executable and as the regular mkosi.version otherwise.

The script is executed unsandboxed (not a fan of this, but unsure how this should be done before any configuration even exists) and reads stdout to figure out the version.

Maybe that's not necessary? Could we run it just before the configure script?

[NekkoDroid]

Having both mkosi.version and mkosi.version.run leaves the path open to reusing the former as input for the latter, as I described earlier, but if we decide we don't need that, this could be conveniently folded into one and use mkosi.version as a script if it's executable and as the regular mkosi.version otherwise.

I can fold them into one mkosi.version, was writing a separate version of this patch that did just that.

Maybe that's not necessary? Could we run it just before the configure script?

I theoretically might be possible, just it would be a bit bigger of a change since ImageVersion= is inherited to the subimage configs, which can override it if they want to and to my knowledge the %v specifier is expanded also during config parsing.

[behrmann]

I theoretically might be possible, just it would be a bit bigger of a change since ImageVersion= is inherited to the subimage configs, which can override it if they want to and to my knowledge the %v specifier is expanded also during config parsing.

That's true, though thinking about it, that's a problem we already have with configure scripts, since they could potentially change the version. That's probably not a solvable problem, since expansions can't be unmade.

So, yes, for least surprise having the version as early as possible seems reasonable.

[NekkoDroid]

To be fair, this isn't the first user script that is run unsandboxed on the host system, the same is done in config_parse_dict.

mkosi/mkosi/config.py

Lines 899 to 902 in d53f31b

	if os.access(p, os.X_OK):
	new[p.name] = run([p], stdout=subprocess.PIPE, env=os.environ).stdout
	else:
	new[p.name] = p.read_text()

Should I change those lines to use config_path_run_or_read, to have them all use the same codepath? That would remove any trailing newlines though.

[NekkoDroid]

Pushed the cleanup commit from #2905 since I had to add the error handling for when mkosi.version is executable anyway :)

[NekkoDroid]

The CI failures seem unrelated?

[behrmann]

I'm travelling over the weekend. Will have a closer look when I have time, Monday at the latest.

The CI failures for Fedora at least are unrelated. We have issues with the Fedora keys about once every release. Haven't looked at the Arch failures.

[behrmann]

Haven't tested it yet, but looks reasonable to me. A few last nits

[septatrix]

Thanks for implementing this Nekko 🎉