PERF: Timestamp/Timedelta constructors when passed a Timestamp/Timedelta

[jschendel]

The Timestamp constructor's performance could be improved when an existing Timestamp object is passed to it via an isinstance-like check:

In [1]: import pandas as pd; pd.__version__
Out[1]: '0.26.0.dev0+1469.ge817ffff3'

In [2]: ts = pd.Timestamp('2020')

In [3]: def timestamp_isinstance_shortcircuit(ts): 
   ...:     if isinstance(ts, pd.Timestamp): 
   ...:         return ts 
   ...:     return pd.Timestamp(ts) 
   ...:

In [4]: %timeit pd.Timestamp(ts)
849 ns ± 13.2 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)

In [5]: %timeit timestamp_isinstance_shortcircuit(ts)
121 ns ± 0.279 ns per loop (mean ± std. dev. of 7 runs, 10000000 loops each)

Some care is needed in the constructor to check if other arguments have been passed, e.g. tz, where we wouldn't be able to directly return the Timestamp object.

Similar story for the Timedelta constructor (should be done in a separate PR):

In [6]: td = pd.Timedelta('1 day')

In [7]: def timedelta_isinstance_shortcircuit(td): 
   ...:     if isinstance(td, pd.Timedelta): 
   ...:         return td 
   ...:     return pd.Timedelta(td) 
   ...:

In [8]: %timeit pd.Timedelta(td)
800 ns ± 1.35 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)

In [9]: %timeit timedelta_isinstance_shortcircuit(td)
120 ns ± 0.269 ns per loop (mean ± std. dev. of 7 runs, 10000000 loops each)

xref #30520

[AlexKirko]

take

[AlexKirko]

Seems straightforward: just add shortcuts to the constructors, make sure they don't fire when unnecessary, and add tests that checks that if we pass a Timestamp / Timedelta object, then we get the same, unmutated object. I'll create the PRs tomorrow.

[AlexKirko]

Ran into a blocking bug while working on this one and submitted an issue (#30692). Will work on it first to solve the current issue cleanly.