pip install -r requirements.txt runs forever (or takes longer than I bothered to wait)

[grahamgower]

It seems there's a conflict somewhere in the version requirements. Installing the requirements/*.txt separately works fine, which explains why there's no failure during CI. We should (1) find/fix the version conflict and (2) add a CI step that installs from this requirements file.

[grahamgower]

It looks like one of jupyter-book's dependencies (markdown-it-py) wants an older version of attrs. There's a newer version of markdown-it-py that relaxes the requirement for attrs<21, but that version is not installed because other dependencies of jupyter-book (jupytext and myst-parser) pin to an older version of markdown-it-py. The docs build just fine if you force pip to install the latest attrs and the old version of markdown-it-py.

I might just leave this as it is and see if it gets resolved with the next batch of dependabot updates next month.

[molpopgen]

This one is biting me right now, too. I'm using pip-compile to explicitly pin everything. It all came crashing down because of this.

[grahamgower]

Are you suggesting that the demes' requirements.txt is confusing your use of pip-compile in fwdpy11, or that you're getting the same problem (coincidentally) because of also depending on jupyter-book and attrs?

A workaround would be to pin to attrs==20.3.0. Maybe we should do that for a while? This version of attrs gives a deprecation warning with demes (specifically for attr.ib(..., cmp=False)), but otherwise works fine. Attrs 21.x has "undeprecated" the cmp option.

[molpopgen]

I managed to show (I think) that it isn't due to anything in demes. It seems to be this mess w/jupyter-book. I just bumped all the attrs down to 20.3.0 and am running tests for that now.

[molpopgen]

One irritation, which I think you may also have noted, is that pip check fails to find any conflicts once I semi-manually get a working environment in place.

[molpopgen]

executablebooks/jupyter-book#1351

[grahamgower]

Fixed by #321.