Contiguous store with unlim dim bug fix

doc/whats-new.rst

@@ -39,6 +39,9 @@ Enhancements
 Bug fixes
 ~~~~~~~~~
 
+- NetCDF4 output: variables with unlimited dimensions must be chunked (not
+  contiguous) on output. (:issue:`1849`)
+  By `James McCreight <https://github.com/jmccreight>`_.
 - indexing with an empty list creates an object with zero-length axis (:issue:`2882`)
   By `Mayeul d'Avezac <https://github.com/mdavezac>`_.
 - Return correct count for scalar datetime64 arrays (:issue:`2770`)

xarray/backends/netCDF4_.py

@@ -210,6 +210,11 @@ def _extract_nc4_variable_encoding(variable, raise_on_invalid=False,
         if chunks_too_big or changed_shape:
             del encoding['chunksizes']
 
+    var_has_unlim_dim = any(dim in unlimited_dims for dim in variable.dims)
+    if (not raise_on_invalid and var_has_unlim_dim
+            and 'contiguous' in encoding.keys()):
+        del encoding['contiguous']
+
     for k in safe_to_drop:
         if k in encoding:
             del encoding[k]
@@ -445,6 +450,7 @@ def prepare_variable(self, name, variable, check_encoding=False,
         encoding = _extract_nc4_variable_encoding(
             variable, raise_on_invalid=check_encoding,
             unlimited_dims=unlimited_dims)
+
         if name in self.ds.variables:
             nc4_var = self.ds.variables[name]
         else:

xarray/tests/test_backends.py

@@ -3525,6 +3525,11 @@ def test_extract_nc4_variable_encoding(self):
         encoding = _extract_nc4_variable_encoding(var, raise_on_invalid=True)
         assert {'shuffle': True} == encoding
 
+        # Variables with unlim dims must be chunked on output.
+        var = xr.Variable(('x',), [1, 2, 3], {}, {'contiguous': True})
+        encoding = _extract_nc4_variable_encoding(var, unlimited_dims=('x',))
+        assert {} == encoding
+
     def test_extract_h5nc_encoding(self):
         # not supported with h5netcdf (yet)
         var = xr.Variable(('x',), [1, 2, 3], {},