Fix for #18178 and #18187 by changing the concat of empty RangeIndex

[SmokinCaterpillar]

The _concat_rangeindex_same_dtype function now keeps track of the last non-empty RangeIndex to extract the new stop value.

This fixes two issues with concatenating non-empty and empty DataFrames and Series.

Two regression tests were added as well.

• closes BUG: pd.concat raises if called on mixture of empty and non-empty dataframes #18178 and closes BUG: pd.concat returns empty series if called on mixture of empty and non-empty series #18187
• 2 regression tests added
• Submission passes git diff master -u -- "*.py" | flake8 --diff
• 1 whatsnew entry

[SmokinCaterpillar]

As an alternative solution, instead of using

if not len(obj):
    continue

and this new last_non_empty variable,
one could simply replace continue as follows:

if not len(obj):
    from pandas import Int64Index
    return _concat_index_same_dtype(indexes, klass=Int64Index)

However, in this case, the returned new index would always be changed to an integer one in case of a single empty RangeIndex in indexes. Accordingly, some already existing tests would need to be changed as well.

[jreback]

i am ok with the soln

[jreback]

maybe filter out he non empties first? then i think this loop should be ok?

[SmokinCaterpillar]

I made the code a bit nicer by first filtering the empty indexes with a list comprehension

[codecov]

Codecov Report

Merging #18191 into master will increase coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master   #18191      +/-   ##
==========================================
+ Coverage    91.4%    91.4%   +<.01%     
==========================================
  Files         163      163              
  Lines       50064    50065       +1     
==========================================
+ Hits        45763    45764       +1     
  Misses       4301     4301

Flag	Coverage Δ
#multiple	89.21% <100%> (ø)	⬆️
#single	40.35% <33.33%> (-0.01%)	⬇️

Impacted Files	Coverage Δ
pandas/core/dtypes/concat.py	99.14% <100%> (ø)	⬆️

---

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 17e0b13...3082d0e. Read the comment docs.

[jorisvandenbossche]

Thanks!

Side note: the behaviour in case of all empties to take the start/stop/step feels a bit strange to me (why not the first? or convert to empty Int64Index?), but given that this is how it already was before, we can leave that for another PR (if we would want to change that)

[SmokinCaterpillar]

@jorisvandenbossche yes, another pull request for these changes is probably a good idea, especially because IMO this involves changing existing RangeIndex tests.

[jreback]

thanks @SmokinCaterpillar

yeah I find this logic slightly strange. If you'd like to follow up with a PR to clarify would be great!

[SmokinCaterpillar]

Opened a follow up PR #18214 :-)