PERF: check first for recognized scalars in DatetimeArray._validate_scalar

[jorisvandenbossche]

Something small, but most of the time we will be setting with actual scalars, so putting that first improves this case a bit:

arr = pd.date_range("2012-01-01", periods=3).array scalar = arr[0] In [3]: %timeit arr._validate_setitem_value(scalar) 3.04 µs ± 58.5 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each) # <-- master 2.08 µs ± 27.2 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each) # <-- PR

(it gives a noticeable difference in eg fillna benchmarks)

[jorisvandenbossche]

The second commits gives another small improvement (the _assert_tzawareness_compat just before already ensures that self and other have either both a tz or both not):

In [3]: %timeit arr._validate_setitem_value(scalar) 3.04 µs ± 58.5 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each) # <-- master 1.85 µs ± 82 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each) # <-- pR

[jbrockmendel]

LGTM

[jreback]

thanks. at some point if you can add the collected PR numbers to the whatsnew in the perf improvements section.

[jorisvandenbossche]

@jreback they're mostly related to (or most relevant for) ArrayManager, so I should probably start with some actual basic docs about it before starting to mention it in the whatsnew

[jreback]

well u can certainly make a section in the whatsnew to indicate that perf for these have changed