Fix regression in .ix fallback with IntervalIndex

[jorisvandenbossche]

closes #27865

cc @jbrockmendel this is another one from limiting the catched exceptions in https://github.com/pandas-dev/pandas/pull/27223/files#diff-7489e290a23303c4db4f803011ecaf8eR121. But also this time it's again rather a bug in the code that raises this error.

For some reason, DataFrame._get_value for a DataFrame with an IntervalIndex raises an AttributeError. That's a bug in itself which should be solved (and probably is the cause of more bugs in other places where _get_value is used, eg in .at). But will open a separate issue for that.

[jorisvandenbossche]

Opened #27927

[jreback]

really? we haven't patched .ix in many releases. And we are removing it.

[jbrockmendel]

@jorisvandenbossche you work on windows right? can you reproduce the azure failure?

[jorisvandenbossche]

No, sorry, no windows :) (linux)

[jorisvandenbossche]

really? we haven't patched .ix in many releases. And we are removing it.

Then we maybe also shouldn't have tried to clean up the code .. ;)

Anyway, this is just for 0.25.1, in master afterwards we can remove it again (when removing ix).
If the idea is that people can upgrade to 0.25, see all the warnings to fix them, to then be able to upgrade to 1.0, then the their code needs to work to be able to do that.

[jbrockmendel]

Do I understand correctly: there is a bug in IntervalTree and fixing that would make this fix unnecessary?

[TomAugspurger]

Yep. But I think that catching AttributeError here is a bit more conservative for the backport.

[jorisvandenbossche]

Yes, I opened #27927 for that (the bug appears in other places as well, but there it was already failing before, so not a regression)

[jorisvandenbossche]

Any objections to merge this?

[TomAugspurger]

I think we're good. I had to restart the CI earlier for a failed build from the flaky test.

[jorisvandenbossche]

@meeseeksdev backport to 0.25.x