RLS: 2.3

[lithomas1]

Release Date: ???

Features:

• Non-breaking PDEP-14 implementation PDEP-14: Dedicated string data type for pandas 3.0 #58551:
  • TRACKER: new default String dtype (pyarrow-backed, numpy NaN semantics) #54792
• groupby.apply DeprecationWarning -> FutureWarning
• Publish Python 3.13 wheels (done in 2.2.3)
• Numpy 2.1 compatibility (may also require dropping Python 3.9) (done in 2.2.3)

cc @pandas-dev/pandas-core

[mroeschke]

I updated the OP with the items to be included in 2.3 based on the pandas dev meeting notes. I believe the release date will be dependent on when the PDEP-14 is implemented and the pandas test suite passes with future.infer_string

The following PRs may need to be backported to 2.3:

• Dropping Python 3.9: DEPS: Drop Python 3.9 #58238
• Numpy 2.1 compat: COMPAT: Fix numpy 2.1 timedelta * DateOffset #59441
• Python 3.13 build wheel changes: (PRs linked from ENH: Python 3.13 free-threading support #59057)

These backports probably should probably come before/after #59513 cc @WillAyd

[lithomas1]

List of TODOS for infer_string (including xfailed tests) can be found here I think.

https://github.com/search?q=repo%3Apandas-dev%2Fpandas+TODO%28infer_string%29+language%3APython&type=code

[rhshadrach]

There is both a 2.3.0 and a 2.3.x branch, which one should we target for PRs?

[WillAyd]

2.3.x I made the 2.3.0 branch by mistake but can't delete it

[rhshadrach]

@jorisvandenbossche - would you have the ability to delete?

[mroeschke]

I went ahead and deleted the 2.3.0 branch

[jorisvandenbossche]

I do wonder if we should consider doing a 2.2.3 with specifically the Python 3.13 and numpy 2.1 changes. Downstream projects might appreciate that we have compatible wheels relatively soon, and when we will be exactly ready for 2.3 is still a bit unsure.

[lithomas1]

I do wonder if we should consider doing a 2.2.3 with specifically the Python 3.13 and numpy 2.1 changes. Downstream projects might appreciate that we have compatible wheels relatively soon, and when we will be exactly ready for 2.3 is still a bit unsure.

That's a good idea.
If we follow the normal release process for 2.3.0, we'd have to do an RC aswell.
Assuming we are able to release an RC next week somehow, that would only leave 2 weeks for addressing any issues from the RC before Python 3.13 comes out, which is a pretty tight schedule.

Personally, I think it's really important that we get a Python 3.13 compatible release out before 3.13 final is due on October 1st.
https://peps.python.org/pep-0719/

Thoughts @pandas-dev/pandas-core?

[jorisvandenbossche]

I don't think next week is realistic for 2.3.0, based on the current speed of progress (I think end of the month would be my best (optimistic) estimate)

[WillAyd]

I would be OK with another 2.2.3 release if someone had time / interest for it. However, same word of caution as with 2.3 - I think getting the backport branch updated and passing tests is a rather large effort, so don't want to underestimate the time that may take

[mikev-junimo]

Has the performance regression with infer strings mentioned in #59657 been fixed in 2.3.x? Because when I enable this option in 2.2 I also experience a perf regression.

[jorisvandenbossche]

@mikev-junimo good question. I took a quick look, and it seems this is not fixed on 2.3.x, only on main (thanks to #57205, but that is not backported). I have opened a smaller PR to also just fix it for 2.3.x: #60461

[mikev-junimo]

@jorisvandenbossche thanks! I've been wanting to enable it in our env but couldn't due to that bug.