DEPR: ignoring missing labels when indexing on MultiIndex level

[jbrockmendel]

• closes #xxxx
• tests added / passed
• Ensure all linting tests pass, see here for how to run them
• whatsnew entry

There are a bunch of issues that boil down to this weird behavior. I need to track down the full list (@attack68 do you have such a list handy?)

[attack68]

There are a bunch of issues that boil down to this weird behavior. I need to track down the full list (@attack68 do you have such a list handy?)

I haven't got a list but I have a script/notebook that tries to iterate over all possible combinations and display the results. I will try and post it later.

I hold the view that the tests for multiindex loc is quite sparse in terms of patchy parametrised tests, but I think a more robust algorithmic approach to all combinations might be worthwhile. Perhaps a routine for this might develop from the script??

[attack68]

https://github.com/attack68/pandas/tree/multiindex_ipynb
file in root: LocIndexing.ipynb

It gathers and analyses many different combinations of loc slicing.

For a MultiIndex with ['a', 'b', 'c', 'd'] in each level, (i.e. '!' not in any level) here is a summary of some odd series.loc slices, when combining with different index types (duplicated vs unique, and monotonic vs unsorted).