BUG: Index may ignore specified datetime/timedelta dtypes

[sinhrks]

• tests added / passed
• passes git diff upstream/master | flake8 --diff
• whatsnew entry

Index ignores specified dtype if it is datetime64 (normal and tz) or timedelta64. This PR makes it consistent with DatetimeIndex and TimedeltaIndex.

pd.Index([1, 2, 3], dtype='datetime64[ns, US/Eastern]')
# Index([1, 2, 3], dtype='object')
pd.Index([1, 2, 3], dtype='datetime64[ns]')
# Index([1, 2, 3], dtype='object')
pd.Index([1, 2, 3], dtype='timedelta64[ns]')
# Int64Index([1, 2, 3], dtype='int64')

Also, fixed MultiIndex.get_level_values not to pass unnecessary tz and freq.

[codecov-io]

Current coverage is 85.27% (diff: 100%)

Merging #13981 into master will decrease coverage by 0.01%

@@             master     #13981   diff @@
==========================================
  Files           139        139          
  Lines         50206      50205     -1   
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
- Hits          42818      42810     -8   
- Misses         7388       7395     +7   
  Partials          0          0          

Powered by Codecov. Last update 29d9e24...a922aef

[jreback]

thanks!

[jorisvandenbossche]

@sinhrks Is it correct that you introduced the tz keyword in Index in this PR?
Personally I don't think this is needed (it is also not documented).

To be clear, the fact that dtype is no longer ignored in Index is a nice bug fix! My comment is only about the tz keyword.

[jreback]

tz is NOT in the Index signature but you can pass kwargs generally

[sinhrks]

@jorisvandenbossche prior to this PR, passing tz kwd coerces to DatetimeIndex in some cases. This PR makes it consistent.

• https://github.com/pydata/pandas/blob/master/pandas/indexes/base.py#L267

[jorisvandenbossche]

tz was previously passed through to when you already had datetime data, eg:

In [10]: pd.Index([datetime.datetime(2012,1,1), datetime.datetime(2012,1, 2)], tz="Europe/Brussels")
Out[10]: DatetimeIndex(['2012-01-01 00:00:00+01:00', '2012-01-02 00:00:00+01:00'], dtype='datetime64[ns, Europe/Brussels]', freq=None)

But the consequence of this PR is that for example the following now also works:

In [82]: pd.Index([1, 2], tz="Europe/Brussels")
Out[82]: DatetimeIndex(['1970-01-01 00:00:00+01:00', '1970-01-01 00:00:00+01:00'], dtype='datetime64[ns, Europe/Brussels]', freq=None)

I personally don't think this should work like that .. (a bit too much magic).