Fix tokenize with empty attrs

doc/whats-new.rst

@@ -50,6 +50,8 @@ Deprecations
 Bug fixes
 ~~~~~~~~~
 
+- Fix bug where empty attrs would generate inconsistent tokens (:issue:`6970`, :pull:`8101`).
+  By `Mattia Almansi <https://github.com/malmans2>`_.
 - Improved handling of multi-coordinate indexes when updating coordinates, including bug fixes
   (and improved warnings for deprecated features) for pandas multi-indexes (:pull:`8094`).
   By `Benoît Bovy <https://github.com/benbovy>`_.

xarray/core/variable.py

@@ -566,7 +566,7 @@ def __dask_tokenize__(self):
         # around NetCDF and the like
         from dask.base import normalize_token
 
-        return normalize_token((type(self), self._dims, self.data, self._attrs))
+        return normalize_token((type(self), self._dims, self.data, self.attrs))
 
     def __dask_graph__(self):
         if is_duck_dask_array(self._data):

xarray/tests/test_dask.py

@@ -299,6 +299,17 @@ def test_persist(self):
         self.assertLazyAndAllClose(u + 1, v)
         self.assertLazyAndAllClose(u + 1, v2)
 
+    def test_tokenize_empty_attrs(self) -> None:
+        # Issue #6970
+        assert self.eager_var._attrs is None
+        expected = dask.base.tokenize(self.eager_var)
+        assert self.eager_var.attrs == self.eager_var._attrs == {}
+        assert (
+            expected
+            == dask.base.tokenize(self.eager_var)
+            == dask.base.tokenize(self.lazy_var.compute())
+        )
+
     @requires_pint
     def test_tokenize_duck_dask_array(self):
         import pint