BUG: groupby's first/last functions maintain Series rather than convert to numpy array (#15884)

[chrisaycock]

• closes BUG: first() loses the timezone in groupby #15884
• tests added / passed
• passes git diff upstream/master --name-only -- '*.py' | flake8 --diff
• whatsnew entry

[jreback]

@chrisaycock not what I was looking for. The only reason that compat routine is called is because the block aggregator raises. So this will not be performant. I don't think these compat routines are actually called at all (or should be). All of the handling is now done via the cython routines (I *think).

[jreback]

So these compat routines are only called if numeric_only=False is passed. It tries the cython path first. So for datetimes it should be working (but is raising because its being called incorrectly).

[chrisaycock]

@jreback Those compat functions were definitely being called for the test case, which is how I found them. If they aren't supposed to be called, then I'll have to take another crack at that issue.

[jreback]

@chrisaycock not what I mean.

put a breakpoint in _cython_agg_blocks. It is hit first (pun intended). This raises (incorrectly), then the compat routines are called.

[chrisaycock]

@jreback Ok, I see what you mean. The .values is definitely a DatetimeIndex instead of a multi-dimensional array.

[codecov]

Codecov Report

Merging #15885 into master will decrease coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master   #15885      +/-   ##
==========================================
- Coverage   90.97%   90.96%   -0.01%     
==========================================
  Files         145      145              
  Lines       49483    49481       -2     
==========================================
- Hits        45015    45012       -3     
- Misses       4468     4469       +1

Flag	Coverage Δ
#multiple	88.72% <100%> (-0.01%)	⬇️
#single	40.63% <0%> (ø)	⬆️

Impacted Files	Coverage Δ
pandas/core/groupby.py	95.02% <100%> (-0.01%)	⬇️
pandas/core/common.py	90.68% <0%> (-0.35%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update da0523a...c0d29ff. Read the comment docs.