Fixes #2198: Drop chunksizes when only when original_shape is different, not when it isn't found (#2207)

* Fixes #2198: Drop chunksizes when original_shape is different

Before this fix chunksizes was dropped even when
original_shape was not found in encoding

* More direct has_original_shape check

* Fixed typo

* Added test if chunksizes is kept when no original shape

* Fix typo in test name

Co-Authored-By: Deepak Cherian <[email protected]>

* Fix keep_chunksizes_if_no_orignal_shape test by using native open_dataset

* Added entry in whats-new

* Use roundtrip mechanism in chunksizes conservation test

Author: Karel-van-de-Plassche

doc/whats-new.rst

@@ -21,6 +21,10 @@ v0.12.2 (unreleased)
 Enhancements
 ~~~~~~~~~~~~
 
+
+- netCDF chunksizes are now only dropped when original_shape is different,
+  not when it isn't found. (:issue:`2207`)
+  By `Karel van de Plassche <https://github.com/Karel-van-de-Plassche>`_.
 - Enable `@` operator for DataArray. This is equivalent to :py:meth:`DataArray.dot`
   By `Maximilian Roos <https://github.com/max-sixty>`_.
 - Add ``fill_value`` argument for reindex, align, and merge operations

xarray/backends/netCDF4_.py

@@ -206,7 +206,9 @@ def _extract_nc4_variable_encoding(variable, raise_on_invalid=False,
         chunks_too_big = any(
             c > d and dim not in unlimited_dims
             for c, d, dim in zip(chunksizes, variable.shape, variable.dims))
-        changed_shape = encoding.get('original_shape') != variable.shape
+        has_original_shape = 'original_shape' in encoding
+        changed_shape = (has_original_shape and
+                         encoding.get('original_shape') != variable.shape)
         if chunks_too_big or changed_shape:
             del encoding['chunksizes']
 

xarray/tests/test_backends.py

@@ -1134,6 +1134,18 @@ def test_encoding_kwarg_compression(self):
 
         assert ds.x.encoding == {}
 
+    def test_keep_chunksizes_if_no_original_shape(self):
+        ds = Dataset({'x': [1, 2, 3]})
+        chunksizes = (2, )
+        ds.variables['x'].encoding = {
+            'chunksizes': chunksizes
+        }
+
+        with self.roundtrip(ds) as actual:
+            assert_identical(ds, actual)
+            assert_array_equal(ds['x'].encoding['chunksizes'],
+                               actual['x'].encoding['chunksizes'])
+
     def test_encoding_chunksizes_unlimited(self):
         # regression test for GH1225
         ds = Dataset({'x': [1, 2, 3], 'y': ('x', [2, 3, 4])})