I don't love running sudo in the test suite, but I couldn't find a another way to delete these files

I wonder why we need to do this manual removal in the first place. In prepare_fixture_dir we add a finalizer that is supposed to clean up the temp dir altogether. Does that not work here? Does it work in other places where we use it to set up a fixture dir?

But if it is needed, some followup questions:

Can we assume sudo is present? What if it asks for credentials?

Any idea if this is needed on all macOS setups, or just the GH one?

If it's a GH only thing, and perhaps otherwise as well, I suppose we could just not fail on error here if the remove fails and let the error message come through so people can clean up manually.

Let's add a comment with details why it doesn't work without it here.

Maybe consider doing the sudo rm in the prepare_fixture_dir finalizer on darwin, if it is needed?

I wonder why we need to do this manual removal in the first place. In prepare_fixture_dir we add a finalizer that is supposed to clean up the temp dir altogether. Does that not work here? Does it work in other places where we use it to set up a fixture dir?

From what I gathered, the issue is that this is the only place where we create more files and folders inside the fixture_dir. Those extra files are created by other processes and not python, so I think that's what causes the issue. See this note in the removefile() syscall:

Write protected files owned by another process cannot be removed by
removefile(), regardless of the permissions on the directory containing
the file.

I might have just not found the correct magic spell to make these files deletable without sudo

But if it is needed, some followup questions:

Can we assume sudo is present? What if it asks for credentials?

Yeah that's a problem. It seems the default behavior is asking for a password

Any idea if this is needed on all macOS setups, or just the GH one?

I don't know, the man pages seem to imply this will happen for all macOS's, but maybe that's not the real issue. I don't have a physical one to test on.

If it's a GH only thing, and perhaps otherwise as well, I suppose we could just not fail on error here if the remove fails and let the error message come through so people can clean up manually.

I'm not sure what a good way to let the error come through without failing the tests is

Let's add a comment with details why it doesn't work without it here.

Maybe consider doing the sudo rm in the prepare_fixture_dir finalizer on darwin, if it is needed?

prepare_fixture_dir doesn't currently have access to bash, so it can't run sudo. It also isn't needed for all of the uses of it. I can make it accept a bash instance if you think that's better though.

Just to confirm, is --color=yes needed here? We don't seem to add it in Linux envs, but do get color nevertheless.

It's not immediately clear to me why this is needed on macOS and not on others, could we elaborate in a comment?

...after having a look further in the PR, it seems various sorts are added elsewhere as well. That smells a bit as if the collation options would not be properly in effect. Or are they, but just sort differently in macOS nevertheless?

Yeah I'm not sure what's up with that. I just checked and LC_COLLATE seems to be set to C. I'll need to look into that

I wonder why we need to do this manual removal in the first place. In prepare_fixture_dir we add a finalizer that is supposed to clean up the temp dir altogether. Does that not work here? Does it work in other places where we use it to set up a fixture dir?

But if it is needed, some followup questions:

Can we assume sudo is present? What if it asks for credentials?

Any idea if this is needed on all macOS setups, or just the GH one?

If it's a GH only thing, and perhaps otherwise as well, I suppose we could just not fail on error here if the remove fails and let the error message come through so people can clean up manually.

Let's add a comment with details why it doesn't work without it here.

Maybe consider doing the sudo rm in the prepare_fixture_dir finalizer on darwin, if it is needed?

No that strong opinions, but perhaps we could take a look if it would become more messier than it's good for if we'd combine this with test/docker/entrypoint.sh with appropriate conditionals to avoid duplicating a bunch of things.