concat keeps attrs from first variable. (#3637)

* concat keep attrs from first variable.

Fixes #2060
Fixes #2575

Author: dcherian

doc/whats-new.rst

@@ -25,6 +25,9 @@ Breaking changes
 
 New Features
 ~~~~~~~~~~~~
+- :py:func:`xarray.concat` now preserves attributes from the first Variable.
+  (:issue:`2575`, :issue:`2060`, :issue:`1614`)
+  By `Deepak Cherian <https://github.com/dcherian>`_.
 - :py:meth:`Dataset.quantile`, :py:meth:`DataArray.quantile` and ``GroupBy.quantile``
   now work with dask Variables.
   By `Deepak Cherian <https://github.com/dcherian>`_.

xarray/core/concat.py

@@ -93,12 +93,14 @@ def concat(
           those of the first object with that dimension. Indexes for the same
           dimension must have the same size in all objects.
 
-    indexers, mode, concat_over : deprecated
-
     Returns
     -------
     concatenated : type of objs
 
+    Notes
+    -----
+    Each concatenated Variable preserves corresponding ``attrs`` from the first element of ``objs``.
+
     See also
     --------
     merge

xarray/core/variable.py

@@ -1625,8 +1625,9 @@ def concat(cls, variables, dim="concat_dim", positions=None, shortcut=False):
         if not shortcut:
             for var in variables:
                 if var.dims != first_var.dims:
-                    raise ValueError("inconsistent dimensions")
-                utils.remove_incompatible_items(attrs, var.attrs)
+                    raise ValueError(
+                        f"Variable has dimensions {list(var.dims)} but first Variable has dimensions {list(first_var.dims)}"
+                    )
 
         return cls(dims, data, attrs, encoding)
 

xarray/tests/test_concat.py

@@ -462,3 +462,16 @@ def test_concat_join_kwarg(self):
         for join in expected:
             actual = concat([ds1, ds2], join=join, dim="x")
             assert_equal(actual, expected[join].to_array())
+
+
+@pytest.mark.parametrize("attr1", ({"a": {"meta": [10, 20, 30]}}, {"a": [1, 2, 3]}, {}))
+@pytest.mark.parametrize("attr2", ({"a": [1, 2, 3]}, {}))
+def test_concat_attrs_first_variable(attr1, attr2):
+
+    arrs = [
+        DataArray([[1], [2]], dims=["x", "y"], attrs=attr1),
+        DataArray([[3], [4]], dims=["x", "y"], attrs=attr2),
+    ]
+
+    concat_attrs = concat(arrs, "y").attrs
+    assert concat_attrs == attr1

xarray/tests/test_variable.py

@@ -432,7 +432,7 @@ def test_concat(self):
         assert_identical(
             Variable(["b", "a"], np.array([x, y])), Variable.concat((v, w), "b")
         )
-        with raises_regex(ValueError, "inconsistent dimensions"):
+        with raises_regex(ValueError, "Variable has dimensions"):
             Variable.concat([v, Variable(["c"], y)], "b")
         # test indexers
         actual = Variable.concat(
@@ -451,16 +451,12 @@ def test_concat(self):
             Variable.concat([v[:, 0], v[:, 1:]], "x")
 
     def test_concat_attrs(self):
-        # different or conflicting attributes should be removed
+        # always keep attrs from first variable
         v = self.cls("a", np.arange(5), {"foo": "bar"})
         w = self.cls("a", np.ones(5))
         expected = self.cls(
             "a", np.concatenate([np.arange(5), np.ones(5)])
         ).to_base_variable()
-        assert_identical(expected, Variable.concat([v, w], "a"))
-        w.attrs["foo"] = 2
-        assert_identical(expected, Variable.concat([v, w], "a"))
-        w.attrs["foo"] = "bar"
         expected.attrs["foo"] = "bar"
         assert_identical(expected, Variable.concat([v, w], "a"))