Fix saving chunked datasets with zero length dimensions (#5742)

* Added test_save_emptydim for zarr backends, which fails when chunking

Added test that fails when saving to zarr a dataset with a chunked array
that has a dimension of length zero (Issue #5741)

* Load all variables with zero entries before saving to_zarr

This addresses Issue #5741 and allows `test_save_emptydim` to pass. We
get around `to_zarr` not liking dask arrays with zero length dimensions
by giving it numpy arrays, which works for some reason

* Updated whats-new.rst with information about fix for #5741

Co-authored-by: Maximilian Roos <5635139+max-sixty@users.noreply.github.com>

Author: jaicher

doc/whats-new.rst

@@ -68,6 +68,9 @@ Deprecations
 
 Bug fixes
 ~~~~~~~~~
+
+- Fix ZeroDivisionError from saving dask array with empty dimension (:issue: `5741`).
+  By `Joseph K Aicher <https://github.com/jaicher>`_.
 - Fixed performance bug where ``cftime`` import attempted within various core operations if ``cftime`` not
   installed (:pull:`5640`).
   By `Luke Sewell <https://github.com/lusewell>`_

xarray/backends/api.py

@@ -1322,6 +1322,11 @@ def to_zarr(
     See `Dataset.to_zarr` for full API docs.
     """
 
+    # Load empty arrays to avoid bug saving zero length dimensions (Issue #5741)
+    for v in dataset.variables.values():
+        if v.size == 0:
+            v.load()
+
     # expand str and Path arguments
     store = _normalize_path(store)
     chunk_store = _normalize_path(chunk_store)

xarray/tests/test_backends.py

@@ -2186,6 +2186,16 @@ def test_to_zarr_append_compute_false_roundtrip(self):
                 with self.open(store) as actual:
                     assert_identical(xr.concat([ds, ds_to_append], dim="time"), actual)
 
+    @pytest.mark.parametrize("chunk", [False, True])
+    def test_save_emptydim(self, chunk):
+        if chunk and not has_dask:
+            pytest.skip("requires dask")
+        ds = Dataset({"x": (("a", "b"), np.empty((5, 0))), "y": ("a", [1, 2, 5, 8, 9])})
+        if chunk:
+            ds = ds.chunk({})  # chunk dataset to save dask array
+        with self.roundtrip(ds) as ds_reload:
+            assert_identical(ds, ds_reload)
+
     @pytest.mark.parametrize("consolidated", [False, True])
     @pytest.mark.parametrize("compute", [False, True])
     @pytest.mark.parametrize("use_dask", [False, True])