CI: Fix ci flake error

[charlesdong1991]

• xref ENH: Support MultiIndex columns in parquet (#34777) #36305
  came across it in ENH: Support MultiIndex columns in parquet (#34777) #36305

[jreback]

hmm on master, ok thanks

[charlesdong1991]

@jreback I think an issue needs to be created to address this, it is the second time this happened: basically flake8 check doesn't detect imported but unused cases in feature PR, only detect the error once PR being merged to master.

Have seen it in #37923 and in #37703 , so ci checks ran successfully in those feature PRs, but actually they should have detected this unused import.

[jreback]

hmm interesting, cc @MarcoGorelli

[jbrockmendel]

when ive seen this in the past it has involved multiple PRs which independently are each correctly passed by flake8, only causing a problem when both are merged

[MarcoGorelli]

when ive seen this in the past it has involved multiple PRs which independently are each correctly passed by flake8, only causing a problem when both are merged

I think this is the issue - in #37584 I got this exact error during the PR CI

---

IMO this makes the case for always asking to merge master rather than restarting CI jobs manually

[charlesdong1991]

ehh, okay, i thought of this scenario as well, but just feel surprised to come across them twice recently

thanks @MarcoGorelli @jbrockmendel shall I close the issue?

[MarcoGorelli]

I'll just dig through the logs to try to verify that that's actually what happened

Just for reference, the exact same command is run during CI for both PRs and builds from master:

pre-commit run --show-diff-on-failure --color=always --all-files

[charlesdong1991]

i closed #37940

thanks for checking out! @MarcoGorelli

[MarcoGorelli]

OK, got it - 3 days ago, Brock opened #37872 and #37865, both of which removed an instance of Label from pandas/core/indexes/period.py .

#37872 was merged first. Then, a couple of days later, #37865 was merged, without it having been updated with changes from master - so its own CI was still green, but when it was merged, both instances of Label were gone and the CI check turned red

[MarcoGorelli]

Anyway, this looks right and passes relevant checks - merging so the CI checks are fixed