ENH: Timestamp constructor now raises more explanatory error message

[fujiaxiang]

Timestamp constructor now raises explanatory error message when year, month or day is missing

• closes raise more explanatory error when failing initialising pd.Timestamp #31200
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

[fujiaxiang]

The error message of datetime.datetime() seems to depend on its version. On my computer I have TypeError: function missing required argument 'year' (pos 1), and in some of the CI environment as well as in @mroeschke 's comment in the original issue (#31200) the error is TypeError: Required argument 'month' (pos 2) not found. For compatibility I decided to use try-except block to extract the message from datetime.datetime() in the test.

The use of re.escape to match exact string in pytest is described in their official documentation.

[jbrockmendel]

@fujiaxiang do you have a measure on how performance is affected? We don't call the constructor this way often, so a small hit is OK.

[fujiaxiang]

@fujiaxiang do you have a measure on how performance is affected? We don't call the constructor this way often, so a small hit is OK.

@jbrockmendel I ran %timeit on current master and this implementation, this is what I get:

# on current master
>>>%timeit pd.Timestamp(year=2020, month=1, day=1)
1.19 µs ± 4.74 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)

# re-built from this PR
>>>%timeit pd.Timestamp(year=2020, month=1, day=1)
1.7 µs ± 2.21 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)

Looks like it takes about 43% longer with this PR.
Let me know if you guys need a more accurate way of performance testing.

[jreback]

thanks