0.16.0 release

[dcherian]

It'd be nice to issue a release soon. We should decide if this is a minor 0.15.2 or major 0.16.0

Please edit this list as you wish.

Must-have

• Support multiple dimensions in DataArray.argmin() and DataArray.argmax() methods #3936 multiple dims argmin, argmax

• Sparse DataArray indexing gives incorrect results #4019, Fix conversion of multiindexed pandas objects to sparse xarray objects #4088; MultiIndex to sparse DataArray bug

• FIX: correct dask array handling in _calc_idxminmax #3922 fix dask array handling in idxmax, idxmin

• Incoherencies between docs in open_mfdataset and combine_by_coords and its behaviour. #4009, Fix 4009 #4173 combine_attrs with open_mfdataset

• Transpose coords by default #3824 transpose coords (if major release)

• Remove old auto combine #3926 remove old autocombine

• Correct dask handling for 1D idxmax/min on ND data #4135 nD dask arrays in idxmax/min

Nice to have

• to_unstacked_dataset broken for single-dim variables #4049, Fix to_unstacked_dataset for single dimension variables. #4094 unstacking merge bug
• Fix/apply ufunc meta dtype #4022 apply_ufunc meta vectorize
• Coordinates passed to interp have nan values #3924 interpolating to coordinates with nans
• sel along 1D non-index coordinates #3925 sel along unindexed 1d coordinates
• dim -> coord in DataArray.integrate #3993 replace dim with coord in integrate
• support for units with pint #3594 unit support with pint
• xarray.open_mzar: open multiple zarr files (in parallel) #4003 mfdataset with zarr
• Limit length of dataarray reprs #3905 length of dataarray reprs
• provide a error summary for assert_allclose #3847 error summary for assert_allclose
• xr.infer_freq #4033 infer_freq
• allow multiindex levels in plots #3938 plotting with multiindex
• Add template xarray object kwarg to map_blocks #3816 map_blocks template
• Proposal for better error message about in-place operation #3976 in-place replacement error message
• pint support for Dataset #3975 pint dataset
• Add NetCDF3 dtype coercion for unsigned integer types #4018 netdf3 dtype coercion
• xarray=1.15.1 regression: Groupby drop multi-index #3985 groupby multi-index bug (ref discussion below & in the issue)

[mathause]

• Fix bool weights #4075 passing boolean weights to weighted mean
• ENH: use dask.array.apply_gufunc in xr.apply_ufunc  #4060 dask.array.apply_gufunc in xr.apply_ufunc (supersedes Fix/apply ufunc meta dtype #4022)
• Fix: apply_ufunc with exclude_dims and vectorize #4130 apply_ufunc with vectorize and exclude_dims

[max-sixty]

As discussed on the dev call, I'll take this one.

I think we need to do 0.16.0 as we had some backward-incompat changes (I can't remember now what they were; but can hunt them down if anyone otherwise has a preference for 0.15.2?)

[dcherian]

Yes we have some breaking changes so definitely 0.16.0

IMO we should do #4184 #4175 and #4163 and then release. #3985 is still waiting for a minimal example so perhaps we should leave that for the next release.

[max-sixty]

Agree! I'll keep an eye on those this week.

[max-sixty]

Thanks for updating @dcherian . Is it worth waiting for #3985? I'm still not sure exactly what the bug is there. I'd vote to leave it until we have a repro.

I was thinking of sneaking in #3891 but didn't want to bring it up as probably won't be able to; definitely not worth waiting on me for!

[dcherian]

I'd vote to leave it until we have a repro.

👍 I'm also not sure what the bug is. In fact it's possible that a bug was fixed and a workaround is what's raising the error now :P

Re: #3891 do you mean set keep_attrs=True by default?

[max-sixty]

Re: #3891 do you mean set keep_attrs=True by default?

Exactly, so this will push to 0.17

[max-sixty]

FYI I will do this towards the end of this week or at the latest this weekend

[dcherian]

Thanks @max-sixty !