Pylint/Astroid gives "RecursionError: maximum recursion depth exceeded" in CI tests

[benjaminysmith]

Recent discussion in PR review: #314 (comment)

Example of error: https://jenkins-ci-prod-01.delphi.cmu.edu/blue/organizations/jenkins/covidcast-indicators/detail/PR-314/3/pipeline/

Discussion on previous PR: #272 (comment)

[chinandrew]

Looks to be an open issue pylint-dev/pylint#3836. Been banging my head on this for a while to no avail, including upgrading astroid directly from their github, trying to align package versions, and changing the python version.

I've been able to reproduce this locally by pulling the pip list from the CI and intsalling those locally, so going to try to narrow down the suspect packages.

[chinandrew]

This appears to be due to pandas and the ubuntu version. Locally downgrading to 1.1.1 makes things works, 1.1.2 or above breaks it. However when I pin this on the CI it didn't work :/. The pip lists locally and on CI were identical as was the python major/minor version (which turns out didn't matter). So then I learned ubuntu-latest on Github Actions is ubuntu 18.04, not 20.04 which is what I'm running. Upgraded that and now things are working. @korlaxxalrok FYI

[sgsmob]

It looks like df = pd.merge(df, other) is a common cause of this problem and replacing it with the equivalent df = df.merge(other) removes the error. Going through and seeing if this is sufficient to at least stop the issue in the near-term.

[chinandrew]

It looks like df = pd.merge(df, other) is a common cause of this problem and replacing it with the equivalent df = df.merge(other) removes the error. Going through and seeing if this is sufficient to at least stop the issue in the near-term.

That's interesting, thanks for looking into it 👍

[chinandrew]

As another confusing data point, I just hit this error on the covidcast repo where the linter didn't complain on a pd.merge(x,y), but then complained when i moved that line into a new file..... 😠