Zarr write + Zarr groups

Changelog.rst

@@ -1,10 +1,16 @@
 Version NEXTVERSION
--------------------
+--------------
 
-**2025-12-??**
+**2026-01-??**
 
+* Write Zarr v3 datasets with `cf.write`, and allow the reading of
+  grouped Zarr v2 and v3 datasets with `cf.read`
+  (https://github.com/NCAS-CMS/cf-python/issues/895)
+* Read Zarr v2 and v3 datasets that contain a group hierarchy with
+  `cf.read` (https://github.com/NCAS-CMS/cf-python/issues/894)
 * Reduce the time taken to import `cf`
   (https://github.com/NCAS-CMS/cf-python/issues/902)
+* New optional dependency: ``zarr>=3.1.3``
 
 ----
 

README.md

@@ -87,7 +87,8 @@ of its array manipulation and can:
 * read field constructs from netCDF, CDL, Zarr, PP and UM datasets with a
   choice of netCDF backends,and in local, http, and s3 locations,
 * create new field constructs in memory,
-* write and append field and domain constructs to netCDF datasets on disk,
+* write and append field and domain constructs to netCDF and Zarr v3
+  datasets on disk,
 * read, create, and manipulate UGRID mesh topologies,
 * read, write, and create coordinates defined by geometry cells,
 * read netCDF and CDL datasets containing hierarchical groups,

cf/functions.py

@@ -3184,7 +3184,7 @@ def environment(display=True, paths=True):
     netCDF4: 1.7.2 /home/miniconda3/lib/python3.12/site-packages/netCDF4/__init__.py
     h5netcdf: 1.3.0 /home/miniconda3/lib/python3.12/site-packages/h5netcdf/__init__.py
     h5py: 3.12.1 /home/miniconda3/lib/python3.12/site-packages/h5py/__init__.py
-    zarr: 3.0.8 /home/miniconda3/lib/python3.12/site-packages/zarr/__init__.py
+    zarr: 3.1.3 /home/miniconda3/lib/python3.12/site-packages/zarr/__init__.py
     s3fs: 2024.12.0 /home/miniconda3/lib/python3.12/site-packages/s3fs/__init__.py
     scipy: 1.15.1 /home/miniconda3/lib/python3.12/site-packages/scipy/__init__.py
     dask: 2025.5.1 /home/miniconda3/lib/python3.12/site-packages/dask/__init__.py
@@ -3210,7 +3210,7 @@ def environment(display=True, paths=True):
     netCDF4: 1.7.2
     h5netcdf: 1.3.0
     h5py: 3.12.1
-    zarr: 3.0.8
+    zarr: 3.1.3
     s3fs: 2024.12.0
     scipy: 1.15.1
     dask: 2025.5.1

cf/read_write/read.py

@@ -316,6 +316,10 @@ class read(cfdm.read):
 
             .. versionadded:: 3.17.0
 
+        {{read store_dataset_shards: `bool`, optional}}
+
+            .. versionadded:: NEXTVERSION
+
         {{read cfa: `dict`, optional}}
 
             .. versionadded:: 3.15.0
@@ -328,6 +332,10 @@ class read(cfdm.read):
 
             .. versionadded:: 3.17.0
 
+        {{read group_dimension_search: `str`, optional}}
+
+            .. versionadded:: (cfdm) NEXTVERSION
+
         umversion: deprecated at version 3.0.0
             Use the *um* parameter instead.
 
@@ -434,6 +442,7 @@ def __new__(
         warn_valid=False,
         dask_chunks="storage-aligned",
         store_dataset_chunks=True,
+        store_dataset_shards=True,
         domain=False,
         cfa=None,
         cfa_write=None,
@@ -445,6 +454,7 @@ def __new__(
         ignore_read_error=False,
         fmt=None,
         file_type=None,
+        group_dimension_search="closest_ancestor",
     ):
         """Read field or domain constructs from a dataset."""
         kwargs = locals()

cf/read_write/um/umread.py

@@ -3575,7 +3575,7 @@ def read(
                 # Return now if there are valid file types
                 return []
 
-        f = self.file_open(filename, parse=True)
+        f = self.dataset_open(filename, parse=True)
 
         info = is_log_level_info(logger)
 
@@ -3598,7 +3598,7 @@ def read(
             for var in f.vars
         ]
 
-        self.file_close()
+        self.dataset_close()
 
         return [field for x in um for field in x.fields if field]
 
@@ -3632,7 +3632,7 @@ def _open_um_file(
                 The open PP or FF file object.
 
         """
-        self.file_close()
+        self.dataset_close()
         try:
             f = File(
                 filename,
@@ -3678,15 +3678,15 @@ def is_um_file(self, filename):
         try:
             # Note: No need to completely parse the file to ascertain
             #       if it's PP or FF.
-            self.file_open(filename, parse=False)
+            self.dataset_open(filename, parse=False)
         except Exception:
-            self.file_close()
+            self.dataset_close()
             return False
         else:
-            self.file_close()
+            self.dataset_close()
             return True
 
-    def file_close(self):
+    def dataset_close(self):
         """Close the file that has been read.
 
         :Returns:
@@ -3700,7 +3700,7 @@ def file_close(self):
 
         self._um_file = None
 
-    def file_open(self, filename, parse=True):
+    def dataset_open(self, filename, parse=True):
         """Open the file for reading.
 
         :Paramters: