More support for missing_value. (#2973)

* More support for missing_value.

Fixes #2871

* lint fixes.

* Use not equivalent instead of not equals check.

* lint fix.

* if → elif so we don't call fillna twice

* Better fix.

Author: dcherian

doc/whats-new.rst

@@ -72,6 +72,8 @@ Bug fixes
   By `Deepak Cherian <https://github.com/dcherian`_.
 - A deep copy deep-copies the coords (:issue:`1463`)
   By `Martin Pletcher <https://github.com/pletchm>`_.
+- Increased support for `missing_value` (:issue:`2871`)
+  By `Deepak Cherian <https://github.com/dcherian>`_.
 - Removed usages of `pytest.config`, which is deprecated (:issue:`2988`)
   By `Maximilian Roos <https://github.com/max-sixty>`_.
 - Fixed performance issues with cftime installed (:issue:`3000`)

xarray/coding/variables.py

@@ -8,6 +8,7 @@
 
 from ..core import dtypes, duck_array_ops, indexing
 from ..core.pycompat import dask_array_type
+from ..core.utils import equivalent
 from ..core.variable import Variable
 
 
@@ -145,11 +146,24 @@ class CFMaskCoder(VariableCoder):
     def encode(self, variable, name=None):
         dims, data, attrs, encoding = unpack_for_encoding(variable)
 
-        if encoding.get('_FillValue') is not None:
+        fv = encoding.get('_FillValue')
+        mv = encoding.get('missing_value')
+
+        if fv is not None and mv is not None and not equivalent(fv, mv):
+            raise ValueError("Variable {!r} has multiple fill values {}. "
+                             "Cannot encode data. "
+                             .format(name, [fv, mv]))
+
+        if fv is not None:
             fill_value = pop_to(encoding, attrs, '_FillValue', name=name)
             if not pd.isnull(fill_value):
                 data = duck_array_ops.fillna(data, fill_value)
 
+        if mv is not None:
+            fill_value = pop_to(encoding, attrs, 'missing_value', name=name)
+            if not pd.isnull(fill_value) and fv is None:
+                data = duck_array_ops.fillna(data, fill_value)
+
         return Variable(dims, data, attrs, encoding)
 
     def decode(self, variable, name=None):

xarray/conventions.py

@@ -82,7 +82,8 @@ def maybe_encode_nonstring_dtype(var, name=None):
         if dtype != var.dtype:
             if np.issubdtype(dtype, np.integer):
                 if (np.issubdtype(var.dtype, np.floating) and
-                        '_FillValue' not in var.attrs):
+                        '_FillValue' not in var.attrs and
+                        'missing_value' not in var.attrs):
                     warnings.warn('saving variable %s with floating '
                                   'point data as an integer dtype without '
                                   'any _FillValue to use for NaNs' % name,

xarray/tests/test_coding.py

@@ -6,7 +6,7 @@
 import xarray as xr
 from xarray.coding import variables
 
-from . import assert_identical, requires_dask
+from . import assert_equal, assert_identical, requires_dask
 
 with suppress(ImportError):
     import dask.array as da
@@ -20,6 +20,23 @@ def test_CFMaskCoder_decode():
     assert_identical(expected, encoded)
 
 
+def test_CFMaskCoder_missing_value():
+    expected = xr.DataArray(np.array([[26915, 27755, -9999, 27705],
+                                      [25595, -9999, 28315, -9999]]),
+                            dims=['npts', 'ntimes'],
+                            name='tmpk')
+    expected.attrs['missing_value'] = -9999
+
+    decoded = xr.decode_cf(expected.to_dataset())
+    encoded, _ = xr.conventions.cf_encoder(decoded, decoded.attrs)
+
+    assert_equal(encoded['tmpk'], expected.variable)
+
+    decoded.tmpk.encoding['_FillValue'] = -9940
+    with pytest.raises(ValueError):
+        encoded, _ = xr.conventions.cf_encoder(decoded, decoded.attrs)
+
+
 @requires_dask
 def test_CFMaskCoder_decode_dask():
     original = xr.Variable(('x',), [0, -1, 1], {'_FillValue': -1}).chunk()