Added support for numpy.bool_

doc/whats-new.rst

@@ -34,7 +34,8 @@ Deprecations
 
 Bug fixes
 ~~~~~~~~~
-
+- Added support for `numpy.bool_` attributes in roundtrips using `h5netcdf` engine with `invalid_netcdf=True` [which casts `bool`s to `numpy.bool_`] (:issue:`4981`, :pull:`4986`).
+  By `Victor Negîrneac <https://github.com/caenrigen>`_.
 
 Documentation
 ~~~~~~~~~~~~~

xarray/backends/api.py

@@ -201,7 +201,7 @@ def check_name(name):
 
 def _validate_attrs(dataset):
     """`attrs` must have a string key and a value which is either: a number,
-    a string, an ndarray or a list/tuple of numbers/strings.
+    a string, an ndarray, a bool/numpy.bool_, or a list/tuple of numbers/strings.
     """
 
     def check_attr(name, value):
@@ -218,10 +218,12 @@ def check_attr(name, value):
                 "serialization to netCDF files"
             )
 
-        if not isinstance(value, (str, Number, np.ndarray, np.number, list, tuple)):
+        if not isinstance(
+            value, (str, Number, np.ndarray, np.number, np.bool_, list, tuple)
+        ):
             raise TypeError(
                 f"Invalid value for attr {name!r}: {value!r} must be a number, "
-                "a string, an ndarray or a list/tuple of "
+                "a string, an ndarray, a bool/numpy.bool_, or a list/tuple of "
                 "numbers/strings for serialization to netCDF "
                 "files"
             )

xarray/tests/test_backends.py

@@ -2541,6 +2541,14 @@ def test_complex(self, invalid_netcdf, warntype, num_warns):
 
         assert recorded_num_warns == num_warns
 
+    def test_numpy_bool_(self):
+        # h5netcdf loads booleans as numpy.bool_, this type needs to be supported
+        # when writing invalid_netcdf datasets in order to support a roundtrip
+        expected = Dataset({"x": ("y", np.ones(5), {"numpy_bool": np.bool_(True)})})
+        save_kwargs = {"invalid_netcdf": True}
+        with self.roundtrip(expected, save_kwargs=save_kwargs) as actual:
+            assert_identical(expected, actual)
+
     def test_cross_engine_read_write_netcdf4(self):
         # Drop dim3, because its labels include strings. These appear to be
         # not properly read with python-netCDF4, which converts them into