BUG: DTI/TDI/PI get_indexer_non_unique with incompatible dtype

[jbrockmendel]

I'm pretty sure we can share more code here, but PeriodEngine needs some work first.

[jreback]

lgtm. needs a rebase. is this user visible at a high level? e.g. in .loc or someting?

[jbrockmendel]

is this user visible at a high level? e.g. in .loc or someting?

Yes, will update the test introduced in #30446

[jreback]

also needs a rebase

[jbrockmendel]

hmm this should probably raise a KeyError instead of a TypeError?

df = pd.DataFrame(
    np.random.randn(5, 3),
    columns=["a", "b", "c"],
    index=pd.date_range("2012", freq="H", periods=5),
)

df = df.iloc[[0, 2, 2, 3]].copy() 

dti = df.index
tdi = pd.TimedeltaIndex(dti.asi8)  # matching i8 values

with pytest.raises(TypeError, match="is not convertible to datetime"):
    df.loc[tdi]

with pytest.raises(TypeError, match="is not convertible to datetime"):
    df["a"].loc[tdi]

[jreback]

hmm this should probably raise a KeyError instead of a TypeError?

df = pd.DataFrame(
    np.random.randn(5, 3),
    columns=["a", "b", "c"],
    index=pd.date_range("2012", freq="H", periods=5),
)

df = df.iloc[[0, 2, 2, 3]].copy() 

dti = df.index
tdi = pd.TimedeltaIndex(dti.asi8)  # matching i8 values

with pytest.raises(TypeError, match="is not convertible to datetime"):
    df.loc[tdi]

with pytest.raises(TypeError, match="is not convertible to datetime"):
    df["a"].loc[tdi]

hmm, yep sounds right

[jreback]

i would test the reverse as well (e.g. TDI is the index with a DTI as indexer). This should also raise for Period, I think? (these can be handled as followons)

[jreback]

thanks