Conda package with an incorrect subdir makes repo unreadable.

[calebzulawski]

I think this is the line in question:

libsolv/ext/repo_conda.c

Lines 314 to 319 in 5dc4f81

	/* if we have a global subdir make sure that it matches */
	if (subdir && pd->subdir && strcmp(subdir, pd->subdir) != 0)
	{
	pd->error = "subdir mismatch";
	return JP_ERROR;
	}

In my situation, I have a custom subdir linux-64-foo, but one broken package has the subdir listed as linux-64. This package makes the entire repo unreadable. I think instead of a hard error, this should be a warning so the repo can still be used.

Also see mamba-org/mamba#2363

[mlschroe]

Yes, we certainly should not return an error here. I think conda just ignores the package subdir in this case. Is that correct?

[wolfv]

I think the other way around, it would ignore the global subdir and use the package subdir.

[mlschroe]

Yeah, that would make more sense. However the code in subdir_data.py contains

meta_in_common = {  # just need to make this once, then apply with .update()
            'arch': repodata.get('info', {}).get('arch'),
            'channel': self.channel,
            'platform': repodata.get('info', {}).get('platform'),
            'schannel': schannel,
            'subdir': subdir,
        }

...

info.update(meta_in_common)

I think that this means it overwrites the subdir from the package.

[mlschroe]

Oh, see issue #502

[calebzulawski]

Just curious, what happens if they're (accidentally) not equal and the package subdir overrides the global subdir? In my situation, and others I believe, the package is actually in the wrong subdirectory, so it won't be possible to download it from the subdir indicated in the package either.

[jpedrick]

It would be great if one way or another this would either skip the package and print an error, or overwrite the subdir for the package. Returning an error here ruins a lot of people's days until the remote repo can be fixed.

The conda classic solver works, so it would make sense to replicate whatever they do even if it's wrong by some POV.

Personally, I would expect and want a warning and the mismatched packages skipped as that would force the mismatch to be noticed/fixed, but wouldn't disrupt users who are uninterested in the broken package.

[mlschroe]

I changed the code so that it overwrites the package subdir with the info subdir.