Bug with nextstrain update in >1.7 about cli HOT 3 CLOSED

Ah, thanks for this. I'll fix it up when I have a chance. It looks like loading the config should guard against the file missing and return a default instead.

from cli.

Hmm. I can't reproduce this on 1.7.1, and indeed the code already guards against the cases I thought you were hitting: https://github.com/nextstrain/cli/blob/master/nextstrain/cli/config.py#L36-L41

That error is what I'd expect if the /Users/trvrb/.nextstrain directory didn't exist by the time path.open was called, which would only happen if the mkdir() silently failed. I can't get it to do so.

from cli.

Ah ha! I think you're hitting this Python bug in path.resolve(), which was fixed first in 3.6.2 (you have 3.6.1) and noted with the following changelog entry:

path.resolve(strict=False) no longer cuts the path after the first element not present in the filesystem. Patch by Antoine Pietri.

Note that strict=False is the default.

Indeed, this description of the effect of the bug matches what we're seeing:

For what it's worth, this bug was the cause of an important downtime in our organization, because someone needed to normalize a path that was later passed to a .mkdir(parents=True), and it was, in some cases, silently doing the mkdir of a wrong path instead of creating all the parents.

This means that the mkdir() in our code on your computer was happening against /Users/trvrb instead of /Users/trvrb/.nextstrain! which matches my diagnosis above.

I'll add a workaround in our code next week. In the meantime, you can mkdir ~/.nextstrain and move on with your life if you haven't already.

On top of all this, I also now see that the default behaviour for path.resolve() changed between 3.5 and 3.6, with no way at all to ask for 3.6's default behaviour (which we want) on 3.5. I'd thought the defaults matched. I'll address that as well.

from cli.