BUG: isnull doesnt handle PeriodNaT properly

[sinhrks]

Related to #7485, isnull and notnull is required to be changed.

[jreback]

hmm...this import is a problem, maybe (hacky), check for attributes freq and ordinal?

[jreback]

I think you can simply cheat hear and do:

return hasattr(val,'ordinal') and val.ordinal == iNaT

and you can make this inline bint as well

[jreback]

see my comments, maybe run a perf check to see if the is_periodnat is an issue

[sinhrks]

Thanks. I've ran perf after changing your point, and found some string related methods take 2 times slower than latest commit. Maybe every call of is_periodnat from checknullobj is problematic, I'll consider better approach.

[jreback]

ok, maybe run with those problematic tests again, they are somewhat random (but the import may be a problem).lmk

[jreback]

pushing, if you get too it in next day or 2 pls update

[jreback]

@sinhrks perf on this?

[jreback]

@sinhrks perf on this change?

[jreback]

@sinhrks status?

[jreback]

@sinhrks status?

[jreback]

@sinhrks ?

[jreback]

@sinhrks can you rebase / revisit

[jreback]

closing pls reopen if/when updated

[sinhrks]

Yep, this is mostly solved by #9134. There is still an issue in object dtype, but currently I don't have good idea to handle it properly without affecting to performance (and I think below example is not popular case...).

idx = pd.Index(['x', pd.Period('NaT', freq='M')])
pd.isnull(idx)
# [False False]