Update makedirs function to not supress OSErrors

[SkyLeite]

The current implementation of makedirs suppresses every OSError possible and assumes the directory already exists when one is raised. This is problematic as there are many reasons why the error could occur, and suppressing it could make it hard to trace the problem back in the future.

My implementation fixes this by checking whether the path exists before attempting to create it, and logs every other OSError appropriately.

I also took the liberty of changing a stray os.makedirs call to our makedirs function.

[aeikum]

Looks reasonable, thanks. Will merge.

[SkyLeite]

Updated a different occurrence of this as well. We're also ignoring errors in the set_dir_casefold_bit but I don't know what that function is supposed to do so I'm not touching it for now.

[aeikum]

The logic seems fine, but is listdir and len the best way to determine if a directory is empty? This code isn't terribly perf-critical, but maybe doing something with a scandir iterator would be more efficient? (I'm not a Python expert; maybe the optimizer is smarter than I think?)

[SkyLeite]

From what I could gather from the scandir proposal, that seems to be correct. The latest commit should address that.

[nanonyme]

Use FileExistsError instead of OSError

[nanonyme]

Instead

except OSError as e:
    if e.errno != errno.ENOTEMPTY:
      raise

[SkyLeite]

As far as I could gather checking if the directory is empty before attempting to do the operation is less expensive than letting it throw an exception and catching it. Would we still prefer going with the latter approach?

[nanonyme]

There's no measurable performance difference between the two (although the deletion and seeing whether it fails is most likely faster as it results in less syscalls on average) but with the exception handling race conditions are impossible. That's typically recommended way to handle these things.

[SkyLeite]

Makes sense. I hadn't considered race conditions. I'll make the changes in a bit :)