API: Breaking Changes in 3.0 (without deprecations)

[jbrockmendel]

Discussed on today's call, this issue to discuss potential changes we may want to make in 3.0 that are too big or cumbersome to deprecate.

• offsets.Day DST behavior BUG/API: implement DayDST #44364
• API: return "correct" missing value scalar from Categorical? #29962
• API: Default to stdlib timezone objects instead of pytz #34916
• Index.__eq__(Series) return NotImplemented DEPR: Index == Series should return NotImplemented #36759
• concat with empty or all-NA entries API: value-dependent behaviour in concat with all-NA data #40893, API: concatting of Series/DataFrame - handling (not skipping) of empty objects #39122
• API: Index.__array_wrap__ stop special-casing bool dtype #52676
• Have MultiIndex attributes return tuple instead of FrozenList API: Change MultiIndex.levels/codes to use tuple instead of FrozenList? #53531
• REF: Proposal: change DataFrame.to_excel to output unformatted excel file #54154

[fangchenli]

We could redo the SQL API to add compat for sqlalchemy 1.4/2.0. #40686.
#36893 is also related. We should come up with a solution to solve both issues.

To add to this. In #45660, seems like people have used the sqlite fullback class for other connection types for many years without any issue. If that is actually a popular use case, we probably should officially support it and expand our SQL tests. And this issue overlaps with #36893.

[jbrockmendel]

@mroeschke can you confirm the axis=None box can be checked?

[mroeschke]

Not completely, #50593 did not add axis=None for sum, prod, var, std, sem

[mroeschke]

@jbrockmendel should we push this as breaking changes for 3.0 without deprecations or should we just start a new issue when we're toward the end of the 2.0 cycle?

[jbrockmendel]

im fine re-purposing this

[jbrockmendel]

think the axis=None one is still viable to get for 2.0?

[mroeschke]

Could be possible. I think the ones not addressed take slightly different code paths so not sure if supporting that is straightforward.

IMO it should be not blocking for 2.0