Update whats-new for v2026.09.0 release

doc/whats-new.rst

@@ -6,10 +6,34 @@
 What's New
 ==========
 
-.. _whats-new.2026.07.1:
+.. _whats-new.2026.09.0:
 
-v2026.07.1 (unreleased)
------------------------
+v2026.09.0 (Sep 29, 2026)
+-------------------------
+
+This release adds read support for Zarr V3 rectilinear (variable-sized) chunks,
+Arrow PyCapsule export for :py:class:`DataArray`, a new
+:py:meth:`DataArray.plot.lines` plotting method, :py:class:`DataTree` support
+in :py:func:`apply_ufunc`, and flox-accelerated groupby medians. Bottleneck is
+now disabled by default, the remaining zarr-python 2 compatibility code has been
+removed, and the minimum ``h5netcdf`` version is now 1.8. It also includes many
+bug fixes.
+
+.. warning::
+
+  This is the last xarray release that will support Python 3.11. Future
+  releases will require Python 3.12 or later.
+
+Thanks to the 45 contributors to this release:
+ANIRUDDHA ADAK, Ahmet Kamer Çivi, Albert Yau, Andrew Scherer, Anirban Mandal,
+Aryan Singh K., Asish Kumar, Chandan P, Charles Turner, Deepak Cherian, Devraj
+Pal, Dipak Chaudhari, Eltsefon Mark, Evan Lyall, Illviljan, Joe Hamman, Johnson
+K C, Jules Chéron, Justus Magin, Kropiunig, Marcus Campbell, Mark Harfouche,
+Matt Van Horn, Matthew Rocklin, Matthias Schabel, Michael Niklas, Nick
+Hodgskin, Patrick N. Raanes, Peter Hron, Puneet Dixit, QinXi, Samuel Le
+Meur-Diebolt, Spencer Clark, Stanley C, Stephan Hoyer, Thomas Kluyver, Tom
+Nicholas, dsk0425-sketch, genrichez, hushen, imam, omsatpute61-afk, stepit, whn
+and ✨Sarah Z✨
 
 New Features
 ~~~~~~~~~~~~
@@ -23,12 +47,18 @@ New Features
 
 - Added `PyArrowCapsule interface <https://arrow.apache.org/docs/format/CDataInterface/PyCapsuleInterface.html#arrow-pycapsule-interface>`_
   to :py:class:`DataArray` (``__arrow_c_schema__`` and ``__arrow_c_stream__``), enabling near zero-copy
-  export to pyarrow, polars or duckdb.
+  export to pyarrow, polars or duckdb (:pull:`11338`).
   By `Jules Chéron <https://github.com/jules-ch>`_.
+
 - Added new plot method :py:meth:`DataArray.plot.lines` which allows creating line plots efficiently in
   a similar manner to :py:meth:`DataArray.plot.scatter`, also available for datasets. (:pull:`7173`)
   By `Jimmy Westling <https://github.com/illviljan>`_.
 
+- :py:func:`apply_ufunc` now accepts :py:class:`DataTree` inputs, applying
+  ``func`` to the datasets at each node and returning trees with the same
+  structure (:pull:`11552`).
+  By `QinXi <https://github.com/QinXi-ai>`_.
+
 - The ``h5netcdf`` backend now reports compression and filter settings in the
   variable ``encoding`` consistently with the ``netCDF4`` backend (using
   ``h5netcdf``'s ``Variable.filters()``). This means data compressed with codecs
@@ -40,7 +70,8 @@ New Features
 Breaking Changes
 ~~~~~~~~~~~~~~~~
 - Disable using bottleneck by default, as certain operations are less numerically
-  stable than the equivalent numpy functions.
+  stable than the equivalent numpy functions. Use
+  ``xr.set_options(use_bottleneck=True)`` to opt back in (:pull:`11461`).
   By `Thomas Kluyver <https://github.com/takluyver>`_.
 - All remaining zarr-python 2.x compatibility code has been removed from the
   zarr backend, following the bump of the minimum ``zarr`` version to 3.0.
@@ -59,11 +90,33 @@ Breaking Changes
 
 Deprecations
 ~~~~~~~~~~~~
+- Passing ``keepdims`` to groupby, resample or rolling reductions (e.g.
+  ``ds.rolling(time=12).mean(keepdims=True)``) now emits a ``FutureWarning``.
+  The argument was previously silently ignored or produced unexpected shapes
+  (:issue:`11518`, :pull:`11521`).
+  By `Charles Turner <https://github.com/charles-turner-1>`_.
 
 
 Bug Fixes
 ~~~~~~~~~
 
+- Treat a full ``MultiIndex`` key with tuple-valued levels as scalar selection,
+  so ``.sel`` no longer preserves a length-1 dimension for nested tuple keys
+  that identify a single row (:issue:`11341`, :pull:`11348`).
+- Warn when tuple-style ``DataArray`` coordinates are renamed by explicitly
+  provided dimension names (:issue:`11234`, :pull:`11292`).
+  By `Asish Kumar <https://github.com/officialasishkumar>`_.
+- Preserve non-grouped coordinates in fallback ``groupby`` reductions when grouping
+  by a non-leading dimension reorders the underlying variable dimensions
+  (:issue:`11188`, :pull:`11290`).
+  By `Sarthak <https://github.com/Sarthak160>`_.
+- Fix :py:meth:`Dataset.chunk` and :py:meth:`DataArray.chunk` raising
+  ``ZeroDivisionError`` when using ``"auto"`` chunks on an object that has
+  a zero-length dimension (:pull:`11486`).
+  By `Charles Turner <https://github.com/charles-turner-1>`_.
+- Fix the shape of a :py:class:`~xarray.indexes.CoordinateTransformIndex`-backed
+  coordinate after ``.transpose()`` for non-square arrays (:pull:`11513`).
+  By `Samuel Le Meur-Diebolt <https://github.com/sdiebolt>`_.
 - Fix a bug in the ``scipy`` backend where mixing non-adjacent scalar and
   array indexers in :py:meth:`Dataset.sel` could silently transpose
   dimension sizes when reading from a closed file object
@@ -76,11 +129,12 @@ Bug Fixes
   By `Alexander Kropiunig <https://github.com/Kropiunig>`_.
 - ``min`` and ``max`` of object arrays with ``skipna=False`` now return ``NaN``
   for slices containing missing values, instead of a result that depended on
-  the position of the missing value (:issue:`11501`, :pull:`#11627`).
+  the position of the missing value (:issue:`11501`, :pull:`11627`).
   By `Michael Niklas <https://github.com/headtr1ck>`_.
 - Avoid pandas' deprecated ``Series.values`` when creating a
   :py:class:`~xarray.Variable` from a timezone-aware :py:class:`pandas.Series`
-  (:issue:`11501`, :pull:`#11627`).
+  (:issue:`11501`, :pull:`11627`).
+  By `Michael Niklas <https://github.com/headtr1ck>`_.
 - Fix deadlocks when reading and writing netCDF files with dask at the same
   time. Combined locks now always acquire their locks in the same order, which
   previously depended on memory addresses, and a failed non-blocking acquire,
@@ -103,27 +157,23 @@ Bug Fixes
   index instead of an empty index when the interval direction conflicts with
   the sign of ``step`` (:pull:`11623`).
   By `Ahmet Kamer Çivi <https://github.com/lowgame>`_.
-- Fix async zarr tests using ``wraps`` with ``autospec=True`` on async methods,
-  which caused ``AsyncMock`` objects to leak through instead of real array data
-  (:pull:`11232`).
-  By `Joe Hamman <https://github.com/jhamman>`_.
 - Preserve NumPy ``StringDType`` variables and coordinates in Zarr format 3
   round trips (:issue:`11466`, :pull:`11474`).
   By `stanbot8 <https://github.com/stanbot8>`_.
 - Fixed dask-backed bottleneck rolling reductions declaring a dtype that could
   differ from the dtype returned by the matching numpy-backed bottleneck path,
-  notably ``object`` instead of ``float64`` for boolean inputs.
+  notably ``object`` instead of ``float64`` for boolean inputs (:pull:`11449`).
   By `Matthew Rocklin <https://github.com/mrocklin>`_.
 - Fix :py:func:`~xarray.plot.utils.label_from_attrs` breaking LaTeX axis labels
   when ``textwrap.wrap`` splits the string between adjacent ``$...$`` blocks,
   producing invalid ``$$`` sequences that matplotlib cannot render
-  (:issue:`11452`).
+  (:issue:`11452`, :pull:`11476`).
   By `Gen Richez <https://github.com/genrichez>`_.
 - Fixed :py:meth:`Dataset.stack` raising ``KeyError`` when a stacked dimension
-  has a falsy but valid name such as ``""``, ``False`` or ``0`` (:issue:`9969`).
+  has a falsy but valid name such as ``""``, ``False`` or ``0`` (:issue:`9969`, :pull:`11477`).
   By `JOhnsonKC201 <https://github.com/JOhnsonKC201>`_.
 - :py:func:`polyval` now propagates ``NaN`` for ``NaT`` entries in ``timedelta64``
-  coordinates instead of returning a large sentinel value (:issue:`11462`).
+  coordinates instead of returning a large sentinel value (:issue:`11462`, :pull:`11478`).
   By `Dipak Chaudhari <https://github.com/dchaudhari7177>`_.
 - The zarr backend now writes boolean arrays with native ``bool`` dtype instead
   of converting them to ``int8``. Zarr supports ``bool`` natively, so the
@@ -134,49 +184,49 @@ Bug Fixes
 - No longer emit a ``SerializationWarning`` about a missing ``_FillValue`` when
   encoding a CF coordinate variable (a 1D variable named after its dimension) to
   an integer dtype. CF forbids missing values in coordinate variables, so a
-  ``_FillValue`` is not expected there (:issue:`10305`).
+  ``_FillValue`` is not expected there (:issue:`10305`, :pull:`11524`).
   By `NoiceHax <https://github.com/NoiceHax>`_.
 - Raise an informative ``TypeError`` when a :py:class:`~xarray.Coordinates` object is
   passed as a coordinate value, e.g. ``ds.assign_coords({"x": coords})``, instead
   of silently creating a broken coordinate. Pass the object directly with
-  ``ds.assign_coords(coords)`` (:issue:`10194`).
-  By `NoiceHex <https://github.com/NoiceHax>`_.
+  ``ds.assign_coords(coords)`` (:issue:`10194`, :pull:`11523`).
+  By `NoiceHax <https://github.com/NoiceHax>`_.
 - Fixed two bugs affecting a :py:class:`~xarray.indexes.CoordinateTransformIndex`-backed
   coordinate after ``.transpose()``: :py:meth:`~xarray.indexes.CoordinateTransformIndex.create_variables`
   (used by, e.g., ``.copy()`` and ``.reindex_like()``-based alignment) discarded the
   transposed dims order and silently reverted to the transform's original order; and
   ``xr.align(..., join="exact")`` raised a spurious ``AlignmentError`` for two objects
   sharing an equal multi-dimensional index whose associated coordinate variables
-  simply had a different dims order (:issue:`11530`).
+  simply had a different dims order (:issue:`11530`, :pull:`11532`).
   By `Samuel Le Meur-Diebolt <https://github.com/sdiebolt>`_.
 - :py:meth:`DataArray.to_series` and :py:meth:`Dataset.to_dataframe` no longer
   call ``.todense()`` on ``sparse.COO``-backed variables, which could raise
   ``MemoryError`` for large, genuinely sparse arrays, or (for
   ``Dataset.to_dataframe``) crash outright since ``sparse.COO`` refuses to
   densify implicitly. ``to_series`` now returns only the array's stored
   entries; ``to_dataframe`` indexes by the union of stored entries across all
-  sparse variables sharing the same dims (:issue:`4007`).
+  sparse variables sharing the same dims (:issue:`4007`, :pull:`11528`).
   By `patnr <https://github.com/patnr>`_.
 - Following `pandas-dev/pandas#64793`_, ensure that resampling an array to a
   ``Day`` frequency along a :py:class:`xarray.CFTimeIndex` produces the same
   results as resampling to an equivalent ``Hour`` frequency, including with the
-  use of ``origin`` and ``offset`` options (:pull:`11546`). This effectively
+  use of ``origin`` and ``offset`` options (:pull:`11547`). This effectively
   rolls back the resample-related changes introduced in :pull:`10650`. By
   `Spencer Clark <https://github.com/spencerkclark>`_.
 - Fix regression where accessing an object-dtype index eagerly attempts to
-  import cftime, slowing down operations.
+  import cftime, slowing down operations (:pull:`11558`).
   By `Peter Hron <https://github.com/peterhron>`_.
 - Fixed :py:meth:`DataArray.coarsen()` and :py:meth:`Dataset.coarsen()` raising a
   type error when applying reduction methods, due to the reduction methods being
-  dynamically generated (:issue:`8136`).
+  dynamically generated (:issue:`8136`, :pull:`11556`).
   By `Andrew Scherer <https://github.com/andrew-s28>`_.
 - Fixed a bug that caused rechunking a multi-dimensional cftime array along a
   subset of its dimensions to raise an error (:issue:`11567`, :pull:`11576`).
   By `Spencer Clark <https://github.com/spencerkclark>`_.
 - :py:meth:`Dataset.diff` and :py:meth:`DataArray.diff` now raise a ``ValueError``
   when ``dim`` is not an existing dimension, instead of silently returning the
   object unchanged. This matches the behavior of other methods such as
-  :py:meth:`Dataset.differentiate` and reductions like ``mean`` (:issue:`7748`).
+  :py:meth:`Dataset.differentiate` and reductions like ``mean`` (:issue:`7748`, :pull:`11628`).
   By `imam2004i <https://github.com/imam2004i>`_.
 - Fixed indexing with an empty indexer array. An empty indexer array is now always
   turned into an empty slice for the backend, so that the in-memory part of the
@@ -201,9 +251,21 @@ Documentation
   (:issue:`7517`, :pull:`11618`).
   By `Anirban Mandal <https://github.com/CoderAnirban71>`_.
 
+Performance
+~~~~~~~~~~~
+- Add flox support for :py:meth:`DataArray.groupby().median`,
+  :py:meth:`Dataset.groupby().median`, :py:meth:`DataArray.resample().median`, and
+  :py:meth:`Dataset.resample().median`. This significantly speeds up median reductions
+  when flox is installed by using flox's blockwise implementation, including
+  rechunking when needed. (:issue:`11238`, :pull:`11239`). By `Samuel Le Meur-Diebolt
+  <https://github.com/sdiebolt>`_.
 
 Internal Changes
 ~~~~~~~~~~~~~~~~
+- Fix async zarr tests using ``wraps`` with ``autospec=True`` on async methods,
+  which caused ``AsyncMock`` objects to leak through instead of real array data
+  (:pull:`11232`).
+  By `Joe Hamman <https://github.com/jhamman>`_.
 
 
 .. _whats-new.2026.07.0:
@@ -255,9 +317,6 @@ Deprecations
 Bug Fixes
 ~~~~~~~~~
 
-- Treat a full ``MultiIndex`` key with tuple-valued levels as scalar selection,
-  so ``.sel`` no longer preserves a length-1 dimension for nested tuple keys
-  that identify a single row (:issue:`11341`).
 - :py:meth:`Dataset.drop_encoding` and :py:meth:`DataArray.drop_encoding` no
   longer copy the underlying data, avoiding excessive memory use on large
   datasets (:issue:`11390`, :pull:`11394`).
@@ -457,13 +516,6 @@ Bug Fixes
   By `Emmanuel Ferdman <https://github.com/emmanuel-ferdman>`_.
 - :func:`combine_by_coords` no longer returns an empty dataset when a generator is passed as ``data_objects`` (:issue:`10114`, :pull:`11265`).
   By `Amartya Anand <https://github.com/SurfyPenguin>`_.
-- Warn when tuple-style ``DataArray`` coordinates are renamed by explicitly
-  provided dimension names (:issue:`11234`, :pull:`11292`).
-  By `Asish Kumar <https://github.com/officialasishkumar>`_.
-- Preserve non-grouped coordinates in fallback ``groupby`` reductions when grouping
-  by a non-leading dimension reorders the underlying variable dimensions
-  (:issue:`11188`, :pull:`11290`).
-  By `Sarthak <https://github.com/Sarthak160>`_.
 - Fix h5netcdf backend module detection and ros3 tests (:issue:`11243`, :pull:`11274`).
   By `Kai Mühlbauer <https://github.com/kmuehlbauer>`_.
 
@@ -494,12 +546,6 @@ Internal Changes
   runtime behavior. This enables CI integration for type stub validation and helps
   prevent type annotation regressions (:issue:`11086`).
   By `Kristian Kollsgård <https://github.com/kkollsga>`_.
-- Add flox support for :py:meth:`DataArray.groupby().median`,
-  :py:meth:`Dataset.groupby().median`, :py:meth:`DataArray.resample().median`, and
-  :py:meth:`Dataset.resample().median`. This significantly speeds up median reductions
-  when flox is installed by using flox's blockwise implementation, including
-  rechunking when needed. (:issue:`11238`, :pull:`11239`). By `Samuel Le Meur-Diebolt
-  <https://github.com/sdiebolt>`_.
 
 - Remove ``setup.py`` file (:pull:`11261`).
   By `Nick Hodgskin <https://github.com/VeckoTheGecko>`_.

pyproject.toml

@@ -447,6 +447,8 @@ slowy = "slowy"
 Commun = "Commun"
 Noice = "Noice"
 Hax = "Hax"
+Yau = "Yau"
+whn = "whn"
 
 # Tests
 Ome = "Ome"

xarray/core/options.py

@@ -321,7 +321,7 @@ class set_options:
         netCDF files with ``open_dataset()`` and ``to_netcdf()`` if ``engine``
         is not explicitly specified. May be any permutation or subset of
         ``['netcdf4', 'h5netcdf', 'scipy']``.
-    use_bottleneck : bool, default: True
+    use_bottleneck : bool, default: False
         Whether to use ``bottleneck`` to accelerate 1D reductions and
         1D rolling reduction operations.
     use_flox : bool, default: True