CLN: avoid runtime imports

[jbrockmendel]

Edits to lib.is_scalar are to put slower python-space checks later so they may get short-circuited.

Everything else should be self-explanatory.

[jreback]

any perf differences?

[jbrockmendel]

any perf differences?

PR

In [3]: %timeit is_scalar(None)                                                                                                                                                                                                                   
55.1 ns ± 0.745 ns per loop (mean ± std. dev. of 7 runs, 10000000 loops each)

master

In [3]: %timeit is_scalar(None)                                                                                                                                                                                                                   
588 ns ± 25.7 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)

Presumably anything that is caught by the isinstance(Fraction, Number) would take a hit, but the difference when I pass an int or float is within the margin of error, and it isn't clear what else would be caught by that isinstance check. Also no difference when passing a list, which is what we'd expect.

[TomAugspurger]

LGTM. @jreback any concerns?

[jreback]

i will look at things tomorrow

[jreback]

this is the correct import; we really really don't want to import directly from .lib for things like this, these should always be from pandas.core.dtypes.common.

[jbrockmendel]

this isn't a hill i want to die on, but the dependency structure consideration is that util._* are supposed to be "upstream" of core

[jreback]

we don't enforce that in any way and this is a complete style break.

[jbrockmendel]

ok, will change

[jreback]

cool, ping on green.

[jbrockmendel]

ping

[jreback]

thanks