implement fillna from 24024, with fixes and tests

[jbrockmendel]

cc @jreback @TomAugspurger

couple of issues with fillna needed sorting out

• The DTA version was operating in-place (fixed+tested)
• The TDA version would raise because it wasn't supported in core.missing (fixed+tested)
• The DTA[tz] version would incorrectly re-localize using the existing constructors, i.e. was dependent on the constructor changes in 24024. With the edits here it is correct regardless of whether the constructor is changed.

[jbrockmendel]

kind of weird that Series[timedelta64].fillna(method='pad') works OK in master, will need to track down why it hadn't been supported directly in core.missing.pad_1d

[codecov]

Codecov Report

Merging #24536 into master will decrease coverage by <.01%.
The diff coverage is 12.5%.

@@            Coverage Diff             @@
##           master   #24536      +/-   ##
==========================================
- Coverage   31.88%   31.87%   -0.01%     
==========================================
  Files         166      166              
  Lines       52427    52438      +11     
==========================================
  Hits        16714    16714              
- Misses      35713    35724      +11

Flag	Coverage Δ
#multiple	30.27% <12.5%> (-0.01%)	⬇️
#single	31.87% <12.5%> (-0.01%)	⬇️

Impacted Files	Coverage Δ
pandas/core/arrays/period.py	41.87% <ø> (+1.26%)	⬆️
pandas/core/missing.py	10.25% <0%> (-0.11%)	⬇️
pandas/core/arrays/datetimelike.py	37.11% <14.28%> (-1.25%)	⬇️

---

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 091cfbb...aa9f457. Read the comment docs.

[codecov]

Codecov Report

Merging #24536 into master will increase coverage by 60.44%.
The diff coverage is 100%.

@@             Coverage Diff             @@
##           master   #24536       +/-   ##
===========================================
+ Coverage   31.88%   92.32%   +60.44%     
===========================================
  Files         166      166               
  Lines       52427    52440       +13     
===========================================
+ Hits        16714    48417    +31703     
+ Misses      35713     4023    -31690

Flag	Coverage Δ
#multiple	90.75% <100%> (+60.46%)	⬆️
#single	43% <12.5%> (+11.12%)	⬆️

Impacted Files	Coverage Δ
pandas/core/arrays/period.py	98.52% <ø> (+57.91%)	⬆️
pandas/core/arrays/datetimelike.py	96.61% <100%> (+58.24%)	⬆️
pandas/core/missing.py	92.3% <100%> (+81.94%)	⬆️
pandas/util/_test_decorators.py	90.54% <0%> (+4.05%)	⬆️
pandas/compat/__init__.py	57.91% <0%> (+9.65%)	⬆️
pandas/core/config_init.py	99.24% <0%> (+9.84%)	⬆️
pandas/core/reshape/util.py	100% <0%> (+11.53%)	⬆️
pandas/core/api.py	100% <0%> (+13.33%)	⬆️
pandas/compat/numpy/__init__.py	92.85% <0%> (+14.28%)	⬆️
pandas/io/formats/console.py	74.24% <0%> (+16.66%)	⬆️
... and 128 more

---

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 091cfbb...52e01ae. Read the comment docs.

[jreback]

is something failing? i agree this is probably the soln. but likely these are converted prior to calling this (to i8)

[jbrockmendel]

Presumably; I haven’t looked at the series code to see how it handles this. Enough core.missing mysteries have popped up today I’ll be giving them a close look after the RC

[jreback]

sure

[jbrockmendel]

gentle ping

[jreback]

thanks