maint: bump minimum dependency versions according to policy

ci/policy.yaml

@@ -26,6 +26,7 @@ policy:
     - hypothesis
     - pytz
     - pytest-reportlog
+    - pyarrow # transitive dependency of dask.dataframe, not an xarray dependency
   # these packages don't fail the CI, but will be printed in the report
   ignored_violations:
     - array-api-strict

doc/whats-new.rst

@@ -17,6 +17,71 @@ New Features
 
 Breaking Changes
 ~~~~~~~~~~~~~~~~
+- The minimum versions of some dependencies were changed (see table below).
+  Notably, the minimum ``zarr`` version is now 3.0. Zarr v2 format data is
+  still readable via ``zarr-python`` 3's built-in compatibility layer; however,
+  ``zarr-python`` 2 is no longer a supported dependency.
+  By `Joe Hamman <https://github.com/jhamman>`_.
+
+.. list-table::
+   :header-rows: 1
+   :widths: 30 20 20
+
+   * - Dependency
+     - Old Version
+     - New Version
+   * - boto3
+     - 1.34
+     - 1.37
+   * - cartopy
+     - 0.23
+     - 0.24
+   * - dask-core
+     - 2024.6
+     - 2025.2
+   * - distributed
+     - 2024.6
+     - 2025.2
+   * - flox
+     - 0.9
+     - 0.10
+   * - h5netcdf
+     - 1.4
+     - 1.5
+   * - h5py
+     - 3.11
+     - 3.13
+   * - iris
+     - 3.9
+     - 3.11
+   * - lxml
+     - 5.1
+     - 5.3
+   * - matplotlib-base
+     - 3.8
+     - 3.10
+   * - numba
+     - 0.60
+     - 0.61
+   * - numbagg
+     - 0.8
+     - 0.9
+   * - packaging
+     - 24.1
+     - 24.2
+   * - rasterio
+     - 1.3
+     - 1.4
+   * - scipy
+     - 1.13
+     - 1.15
+   * - toolz
+     - 0.12
+     - 1.0
+   * - zarr
+     - 2.18
+     - 3.0
+
 - Xarray will no longer by default decode a variable into a
   :py:class:`np.timedelta64` dtype based on the presence of a timedelta-like
   ``"units"`` attribute alone. Instead it will rely on the presence of a

pixi.toml

@@ -20,7 +20,7 @@ python = "*"
 numpy = "*"
 pandas = "*"
 
-packaging = "24.1.*"
+packaging = "24.2.*"
 git = "*"            # needed for dynamic versioning
 
 [dependencies]
@@ -107,40 +107,41 @@ numpy = "1.26.*"
 pandas = "2.2.*"
 
 [feature.minimum-scipy.dependencies]
-scipy = "1.13.*"
+scipy = "1.15.*"
 
 [feature.min-versions.dependencies]
 array-api-strict = "2.4.*" # dependency for testing the array api compat
-boto3 = "1.34.*"
+boto3 = "1.37.*"
 bottleneck = "1.4.*"
-cartopy = "0.23.*"
+cartopy = "0.24.*"
 cftime = "1.6.*"
-dask-core = "2024.6.*"
-distributed = "2024.6.*"
-flox = "0.9.*"
-h5netcdf = "1.4.*"
+dask-core = "2025.2.*"
+distributed = "2025.2.*"
+flox = "0.10.*"
+h5netcdf = "1.5.*"
 # h5py and hdf5 tend to cause conflicts
 # for e.g. hdf5 1.12 conflicts with h5py=3.1
 # prioritize bumping other packages instead
-h5py = "3.11.*"
+h5py = "3.13.*"
 hdf5 = "1.14.*"
-iris = "3.9.*"
-lxml = "5.1.*"            # Optional dep of pydap
-matplotlib-base = "3.8.*"
+iris = "3.11.*"
+lxml = "5.3.*"             # Optional dep of pydap
+matplotlib-base = "3.10.*"
 nc-time-axis = "1.4.*"
 # netcdf follows a 1.major.minor[.patch] convention
 # (see https://github.com/Unidata/netcdf4-python/issues/1090)
 netcdf4 = "1.6.*"
-numba = "0.60.*"
-numbagg = "0.8.*"
-packaging = "24.1.*"
+numba = "0.61.*"
+numbagg = "0.9.*"
+packaging = "24.2.*"
 pint = "0.24.*"
 pydap = "3.5.*"
-rasterio = "1.3.*"
+rasterio = "1.4.*"
 seaborn = "0.13.*"
 sparse = "0.15.*"
-toolz = "0.12.*"
-zarr = "2.18.*"
+toolz = "1.0.*"
+zarr = "3.0.*"
+pyarrow = "*"        # required by dask.dataframe
 
 # TODO: Remove `target.unix` restriction once pandas nightly has win-64 wheels again.
 # Without this, `pixi lock` fails because it can't solve the nightly feature for win-64,

pyproject.toml

@@ -18,7 +18,7 @@ name = "xarray"
 readme = "README.md"
 requires-python = ">=3.11"
 
-dependencies = ["numpy>=1.26", "packaging>=24.1", "pandas>=2.2"]
+dependencies = ["numpy>=1.26", "packaging>=24.2", "pandas>=2.2"]
 
 # We don't encode minimum requirements here (though if we can write a script to
 # generate the text from `min_deps_check.py`, that's welcome...). We do add
@@ -27,27 +27,27 @@ dependencies = ["numpy>=1.26", "packaging>=24.1", "pandas>=2.2"]
 
 [project.optional-dependencies]
 accel = [
-  "scipy>=1.13",
+  "scipy>=1.15",
   "bottleneck",
-  "numbagg>=0.8",
+  "numbagg>=0.9",
   "numba>=0.62",  # numba 0.62 added support for numpy 2.3
-  "flox>=0.9",
+  "flox>=0.10",
   "opt_einsum",
 ]
 complete = ["xarray[accel,etc,io,parallel,viz]"]
 io = [
   "netCDF4>=1.6.0",
-  "h5netcdf[h5py]>=1.4.0",
+  "h5netcdf[h5py]>=1.5.0",
   "pydap",
-  "scipy>=1.13",
-  "zarr>=2.18",
+  "scipy>=1.15",
+  "zarr>=3.0",
   "fsspec",
   "cftime",
   "pooch",
 ]
 etc = ["sparse>=0.15"]
 parallel = ["dask[complete]"]
-viz = ["cartopy>=0.23", "matplotlib>=3.8", "nc-time-axis", "seaborn"]
+viz = ["cartopy>=0.24", "matplotlib>=3.10", "nc-time-axis", "seaborn"]
 types = [
   "pandas-stubs",
   "scipy-stubs",

xarray/testing/assertions.py

@@ -45,7 +45,7 @@ def _data_allclose_or_equiv(arr1, arr2, rtol=1e-05, atol=1e-08, decode_bytes=Tru
     if any(arr.dtype.kind == "S" for arr in [arr1, arr2]) and decode_bytes:
         arr1 = _decode_string_data(arr1)
         arr2 = _decode_string_data(arr2)
-    exact_dtypes = ["M", "m", "O", "S", "U"]
+    exact_dtypes = ["M", "m", "O", "S", "U", "T"]
     if any(arr.dtype.kind in exact_dtypes for arr in [arr1, arr2]):
         return duck_array_ops.array_equiv(arr1, arr2)
     else:

xarray/tests/test_backends.py

@@ -524,7 +524,7 @@ def check_dtypes_roundtripped(self, expected, actual):
 
             actual_dtype = actual.variables[k].dtype
             # TODO: check expected behavior for string dtypes more carefully
-            string_kinds = {"O", "S", "U"}
+            string_kinds = {"O", "S", "U", "T"}
             assert expected_dtype == actual_dtype or (
                 expected_dtype.kind in string_kinds
                 and actual_dtype.kind in string_kinds

xarray/tests/test_backends_datatree.py

@@ -704,20 +704,28 @@ def test_zarr_encoding(self, tmpdir, simple_datatree, zarr_format) -> None:
             codec = Blosc(cname="zstd", clevel=3, shuffle=2)
             comp = {"compressors": (codec,)} if has_zarr_v3 else {"compressor": codec}
         elif zarr_format == 3:
-            # specifying codecs in zarr_format=3 requires importing from zarr 3 namespace
-            from zarr.registry import get_codec_class
+            import zarr
 
-            Blosc = get_codec_class("numcodecs.blosc")
-            comp = {"compressors": (Blosc(cname="zstd", clevel=3),)}  # type: ignore[call-arg]
+            comp = {
+                "compressors": (zarr.codecs.BloscCodec(cname="zstd", clevel=3),),
+            }
 
         enc = {"/set2": dict.fromkeys(original_dt["/set2"].dataset.data_vars, comp)}
         original_dt.to_zarr(filepath, encoding=enc, zarr_format=zarr_format)
 
         with open_datatree(filepath, engine="zarr") as roundtrip_dt:
             compressor_key = "compressors" if has_zarr_v3 else "compressor"
-            assert (
-                roundtrip_dt["/set2/a"].encoding[compressor_key] == comp[compressor_key]
-            )
+            if zarr_format == 3:
+                # zarr v3 BloscCodec auto-tunes typesize and shuffle on write,
+                # so we only check the attributes we explicitly set
+                rt_codec = roundtrip_dt["/set2/a"].encoding[compressor_key][0]
+                assert rt_codec.cname.value == "zstd"
+                assert rt_codec.clevel == 3
+            else:
+                assert (
+                    roundtrip_dt["/set2/a"].encoding[compressor_key]
+                    == comp[compressor_key]
+                )
 
             enc["/not/a/group"] = {"foo": "bar"}  # type: ignore[dict-item]
             with pytest.raises(ValueError, match=r"unexpected encoding group.*"):