BUG: first("1M") returning two months when first day is last day of month

[phofl]

• closes BUG: incorrect output of first('1M') in case if first index is the last day of the month #29623
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

[jbrockmendel]

Could we use rollforward?

[phofl]

Instead of the if-else? Then no, This would work for offsets like 1M, but not for something like 10 Days. That is also the thing which lead me to putting offset._day_opt == "end" in. But we could use it like

        if offset._day_opt == "end":
            end_date = end = offset.rollforward(self.index[0])
        else:
            end_date = end = self.index[0] + offset

if this would be preferrably

[jbrockmendel]

this is an OK solution, but ideally there should be a way to do this with only public methods.

but not for something like 10 Days

can you give an example? (possibly merits a test so i dont try to incorrectly simplify this somewhere down the line) if its only for Tick offsets, special-casing wouldn't be that bad since we already special case them a few lines down

[phofl]

Tick works, thanks.

test_first_subset in pandas/tests/frame/methods/test_first_and_last.py covers the behavior and avoids the simplification (as I learned myself a bit earlier :))

[jbrockmendel]

LGTM

[jreback]

thanks @phofl

[jreback]

@meeseeksdev backport 1.2.x

[simonjayhawkins]

@phofl this didn't fix the "2M" case from the issue?

>>> import pandas as pd
>>>
>>> pd.__version__
'1.3.0.dev0+42.g36c4d5ccf1'
>>>
>>> x = pd.Series(1, index=pd.bdate_range("2010-03-31", periods=100))
>>>
>>> print(x.first("1M"))
2010-03-31    1
Freq: B, dtype: int64
>>> print(x.first("2M"))
2010-03-31    1
Freq: B, dtype: int64
>>>

[phofl]

Unfortunately you are right. Opened #38446 to adress this.
Forgot to add the test concerning this case unfortunately