Raise TypeError when Dataset passed as data_vars to Dataset constructor

Dataset(ds) silently dropped attrs because Dataset inherits from Mapping,
so the constructor iterated over data variable keys while ignoring attrs.
This was never intended behavior. Raise TypeError with a message directing
users to ds.copy() instead.

Closes GH11095

Co-authored-by: Claude <noreply@anthropic.com>

Author: kkollsga

doc/whats-new.rst

@@ -28,6 +28,11 @@ Breaking Changes
   not intended to be visible to end-users so this version of xarray
   switches to using ``FutureWarning`` everywhere (:pull:`11112`).
   By `Julia Signell <https://github.com/jsignell>`_.
+- Passing a :py:class:`Dataset` as ``data_vars`` to the :py:class:`Dataset`
+  constructor now raises :py:class:`TypeError`. This was never intended behavior
+  and silently dropped ``attrs``. Use :py:meth:`Dataset.copy` instead
+  (:issue:`11095`).
+  By `Kristian Kollsga <https://github.com/kkollsga>`_.
 
 Deprecations
 ~~~~~~~~~~~~

xarray/core/dataset.py

@@ -385,6 +385,11 @@ def __init__(
     ) -> None:
         if data_vars is None:
             data_vars = {}
+        if isinstance(data_vars, Dataset):
+            raise TypeError(
+                "Passing a Dataset as `data_vars` to the Dataset constructor is"
+                " not supported. Use `ds.copy()` to create a copy of a Dataset."
+            )
         if coords is None:
             coords = {}
 

xarray/tests/test_dataset.py

@@ -494,6 +494,14 @@ def test_constructor(self) -> None:
         actual = Dataset({"z": expected["z"]})
         assert_identical(expected, actual)
 
+    def test_constructor_dataset_as_data_vars_raises(self) -> None:
+        ds = Dataset({"x": ("x", [1, 2, 3])}, attrs={"key": "value"})
+        with pytest.raises(
+            TypeError,
+            match=r"Passing a Dataset as `data_vars`.*Use `ds\.copy\(\)`",
+        ):
+            Dataset(ds)
+
     def test_constructor_1d(self) -> None:
         expected = Dataset({"x": (["x"], 5.0 + np.arange(5))})
         actual = Dataset({"x": 5.0 + np.arange(5)})