Pandas 3 support

[headtr1ck]

What is your issue?

Pandas 3 was released one month ago.

See changelog: https://pandas.pydata.org/docs/whatsnew/v3.0.0.html

I think our upstream testing should have caught most of the issues, except support for the new StringDType.

There is already the first bug report about it: #11199

Maybe we should pin to pandas < 3 until we are certain that we support it fully?

[keewis]

I don't think we should create upper bounds, if at all these should be != bounds. However, even that I don't think is necessary, xarray is not generally unusable with pandas>=3.

Instead, there are some very specific areas of xarray (e.g. encoding or concatenation (#11189, #11183) of index variables with pandas dtypes) that don't yet work properly.

Should we convert this to a meta-issue about pandas 3 support?

[jsignell]

Yeah I agree with @keewis that it's better to just fix issues as they come up rather than do upper bounds (see #11128 (comment)). We have already merged a few pandas 3 related fixes and the regular CI environments are using pandas 3.