RLS: pandas 1.5.3

[datapythonista]

Will be releasing when the blockers are addressed:

• Change iloc inplace setitem FutureWarning to DeprecationWarning BUG: Change FutureWarning to DeprecationWarning for inplace setitem with DataFrame.(i)loc #50044
• Numpy 1.24 bug w/ pivot table w/ mixed nested elements BUG: pivot_table with nested elements and numpy 1.24 #50682

[datapythonista]

Seems like there is not much added to the 1.5.3 milestone, even if we skipped the December release.

Maybe we can release 1.5.3 after the dev call next week as usual, with what's merged, and anything you want to see in 1.5, and plan the release of 2.0 RC for February?

CC: @pandas-dev/pandas-core

[MarcoGorelli]

sounds good!

[Dr-Irv]

seems like there is not much added to the 1.5.3 milestone, even if we skipped the December release.

Not much? If I go to https://github.com/pandas-dev/pandas/milestone/100 it shows 17 open issues and 55 closed ones!

[datapythonista]

Not much? If I go to https://github.com/pandas-dev/pandas/milestone/100 it shows 17 open issues and 55 closed ones!

My bad, you're right. I guess I had some extra filter when I checked, that's not what I saw. Probably worth continuing to 1.5.4 and possibly 1.5.5 for now then, and aim to releasing 2.0 around March or April.

[phofl]

I‘d prefer doing the rc for 2.0 in February anyway (if we get everything in by then). We can still work on 1.5.4 if necessary

[MarcoGorelli]

Yeah TBH I wouldn't block 2.0, it's been 2 years already and none of the regressions look that crucial to me (though I may be wrong), if they don't get fixed by 2.0 they'll get fixed later

[datapythonista]

We always released sequentially until now. I think the releases in the home page, and not sure if something else, will need to be changed to show the latest version as 2.x instead of 1.5.x if 1.5.x is released later. But probably worth doing it anyway, as I guess we'll eventually have a patch release after a major or minor.

Let's release independently then. I'll be releasing 1.5.3 after the dev call, as we've been doing. And we can release 2.0 as soon as it's ready. I see we've got things in the milestone that are probably not blockers. I guess the main things are enforcing deprecations and the copy-on-write stuff, no? Maybe we can add to #46776 the inventory of blocker issues and PRs, so we can better track the status. Or maybe remove from the milestone everything that is not a blocker?

[datapythonista]

I added a list of blockers for 2.0 in #46776. Seems like there is consensus to not delay it until the 1.5.x series is over, and release it once the blockers are addressed.

For 1.5.3, is there any blocker? Assuming there will be at least a 1.5.4? I'm happy to release it early next week if there is nothing in particular to wait for. And then we can focus on 2.0 in the dev call.

[Dr-Irv]

Not sure if #50558 is considered a blocker or not since we're no longer compatible with the latest SQLAlchemy
Also #50342 reveals an issue with numpy 1.24.0. Again, not sure if this is a blocker or not.

[cdcadman]

I don't know if it's possible or not, but I'd be delighted if pandas 1.5.3 could support sqlalchemy 2.0. I have an open PR to get the main branch working with sqlalchemy 2.0 (#48576).

[phofl]

Hm we are only back porting regression fixes in general. The Alchemy changes are too substantial to backport I think

[MarcoGorelli]

if you need this, you should be able to install the nightly pandas wheel, see https://anaconda.org/scipy-wheels-nightly/pandas (and that'd generally be really useful to help pandas catch bugs before the final 2.0 release)

[lithomas1]

I'm +0 on backporting.
Since pandas 2.0 is breaking, some users might be stuck on older pandas (not necessarily by their choice, since some downstream libraries may pin to less than 2.0) for a while, so I would be fine backporting as long as the API is not changed.

[phofl]

I am more weary of introducing new bugs for users that can’t upgrade. Alchemy is a major breaking change with 2.0, so you have to put a lot of effort in to upgrade it as well