Compatibility with Dask 2025.4

docs/src/whatsnew/3.12.rst

@@ -198,6 +198,9 @@ v3.12.1 (30 Apr 2025)
    versions of dask fix the bug casuing the pin. Introduced a minimum pin (2025.1.0)
    to avoid this bug. (:pull:`6342`)
 
+#. `@trexfeathers`_ refactored Iris loading and saving to make it compatible
+   with Dask version ``2025.4.0`` and above. (:pull:`6451`)
+
 
 📚 Documentation
 ================

lib/iris/fileformats/netcdf/saver.py

@@ -1742,7 +1742,7 @@ def _create_generic_cf_array_var(
             if len(element_dims) == 1:
                 data_first = data[0]
                 if is_lazy_data(data_first):
-                    data_first = data_first.compute()
+                    data_first = dask.compute(data_first)
                 data = list("%- *s" % (string_dimension_depth, data_first))
             else:
                 orig_shape = data.shape
@@ -2380,7 +2380,7 @@ def complete(self) -> None:
             raise ValueError(msg)
 
         # Complete the saves now
-        self.delayed_completion().compute()
+        dask.compute(self.delayed_completion())
 
 
 def save(

lib/iris/tests/integration/netcdf/test_delayed_save.py

@@ -9,7 +9,6 @@
 from cf_units import Unit
 import dask.array as da
 import dask.config
-from dask.delayed import Delayed
 import distributed
 import numpy as np
 import pytest
@@ -20,6 +19,13 @@
 from iris.tests.stock import realistic_4d
 
 
+def are_dask_collections(obj):
+    result = dask.is_dask_collection(obj)
+    if not result and hasattr(obj, "__len__"):
+        result = all([dask.is_dask_collection(item) for item in obj])
+    return result
+
+
 class Test__lazy_stream_data:
     # Ensure all saves are done with split-atttribute saving,
     # -- because some of these tests are sensitive to unexpected warnings.
@@ -130,7 +136,7 @@ def test_realfile_loadsave_equivalence(self, save_is_delayed, output_path):
         result = iris.save(original_cubes, output_path, compute=not save_is_delayed)
         if save_is_delayed:
             # In this case, must also "complete" the save.
-            result.compute()
+            dask.compute(result)
         reloaded_cubes = iris.load(output_path)
         reloaded_cubes = sorted(reloaded_cubes, key=lambda cube: cube.name())
         assert reloaded_cubes == original_cubes
@@ -188,7 +194,7 @@ def test_scheduler_types(self, output_path, scheduler_type, save_is_delayed):
         if not save_is_delayed:
             assert result is None
         else:
-            assert isinstance(result, Delayed)
+            assert are_dask_collections(result)
 
     def test_time_of_writing(self, save_is_delayed, output_path, scheduler_type):
         # Check when lazy data is *actually* written :
@@ -251,7 +257,7 @@ def fetch_masks():
 
         if save_is_delayed:
             # Complete the write.
-            result.compute()
+            dask.compute(result)
 
             # Re-fetch the lazy arrays.  The data should now **not be masked**.
             data_mask, coord_mask, ancil_mask, cm_mask = fetch_masks()
@@ -267,4 +273,4 @@ def test_no_delayed_writes(self, output_path):
         cube = self.make_testcube(include_lazy_content=False)
         warnings.simplefilter("error")
         result = iris.save(cube, output_path, compute=False)
-        assert isinstance(result, Delayed)
+        assert are_dask_collections(result)

lib/iris/tests/integration/netcdf/test_general.py

@@ -16,6 +16,7 @@
 from unittest import mock
 import warnings
 
+import dask
 import numpy as np
 import numpy.ma as ma
 import pytest
@@ -456,7 +457,7 @@ def test_basic_save(self):
         ds.close()
 
         # Complete the delayed saves.
-        result.compute()
+        dask.compute(result)
 
         # Check that data now *is* written.
         ds = nc.Dataset(filepath_indirect)

lib/iris/tests/unit/fileformats/pp/test__interpret_field.py

@@ -13,7 +13,6 @@
 
 import numpy as np
 
-import iris
 import iris.fileformats.pp as pp
 
 
@@ -91,43 +90,6 @@ def test_deferred_fix_lbrow_lbnpt(self):
         self.assertEqual(f1.lbrow, 3)
         self.assertEqual(f1.lbnpt, 4)
 
-    @tests.skip_data
-    def test_landsea_unpacking_uses_dask(self):
-        # Ensure that the graph of the (lazy) landsea-masked data contains an
-        # explicit reference to a (lazy) landsea-mask field.
-        # Otherwise its compute() will need to invoke another compute().
-        # See https://github.com/SciTools/iris/issues/3237
-
-        # This is too complex to explore in a mock-ist way, so let's load a
-        # tiny bit of real data ...
-        testfile_path = tests.get_data_path(
-            ["FF", "landsea_masked", "testdata_mini_lsm.ff"]
-        )
-        landsea_mask, soil_temp = iris.load_cubes(
-            testfile_path, ("land_binary_mask", "soil_temperature")
-        )
-
-        # Now check that the soil-temp dask graph correctly references the
-        # landsea mask, in its dask graph.
-        lazy_mask_array = landsea_mask.core_data()
-        lazy_soildata_array = soil_temp.core_data()
-
-        # Work out the main dask key for the mask data, as used by 'compute()'.
-        mask_toplev_key = (lazy_mask_array.name,) + (0,) * lazy_mask_array.ndim
-        # Get the 'main' calculation entry.
-        mask_toplev_item = lazy_mask_array.dask[mask_toplev_key]
-        # This should be a task (a simple fetch).
-        self.assertTrue(callable(mask_toplev_item))
-        # Get the key (name) of the array that it fetches.
-        mask_data_name = mask_toplev_item.args[0].key
-
-        # Check that the item this refers to is a PPDataProxy.
-        self.assertIsInstance(lazy_mask_array.dask[mask_data_name], pp.PPDataProxy)
-
-        # Check that the soil-temp graph references the *same* lazy element,
-        # showing that the mask+data calculation is handled by dask.
-        self.assertIn(mask_data_name, lazy_soildata_array.dask.keys())
-
 
 if __name__ == "__main__":
     tests.main()