PERF: DatetimeIndex.get_indexer with mismatched tz

[jbrockmendel]

Avoid casting to object dtype.

In [2]: dti = pd.date_range("2016-01-01", periods=10000, tz="US/Pacific")
In [3]: dti2 = dti.tz_convert("UTC")

In [4]: %timeit dti.get_indexer(dti2)
127 ms ± 4.69 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)  # <-- master
352 µs ± 10.2 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)  # <-- PR

[jreback]

why should this not raise?

[jbrockmendel]

This is relevant in three places: get_indexer, get_indexer_non_unique, and join. In the first two, we're talking about comparisons, which always allow mixed tz. In the third, that is just The Way It Is, xref #39328

[jreback]

ok i see, yeah generally i find this weird that we allow indexing on mixed tz's but i guess that boat has sailed.

[jreback]

do we have any asv's on user facing, e.g. loc with mixed tz's?

[jbrockmendel]

do we have any asv's on user facing, e.g. loc with mixed tz's?

i dont think so, this implements one that directly measures get_indexer, with a comment that it reached via Series.__getitem__

[jbrockmendel]

do we have any asv's on user facing, e.g. loc with mixed tz's?

i dont think so, this implements one that directly measures get_indexer, with a comment that it reached via Series.__getitem__

[jreback]

thanks