Fix concat failure with pandas StringDtype indexes (#11317)

doc/whats-new.rst

@@ -29,6 +29,10 @@ Bug Fixes
 - Fix a major performance regression in :py:meth:`Coordinates.to_index` (and
   consequently :py:meth:`Dataset.to_dataframe`) caused by converting the cached
   code ndarrays into Python lists (:issue:`11305`).
+- Fix :py:func:`concat` failing with ``TypeError: Cannot interpret
+  '<StringDtype...>' as a data type`` when a ``pandas.Index`` with a
+  ``StringDtype`` is used as the new concat dimension and another input has a
+  numpy string-dtype coord (:issue:`11317`).
 
 
 Documentation

xarray/structure/concat.py

@@ -363,6 +363,13 @@ def _calc_concat_dim_index(
         else:
             (dim,) = dim_or_data.dims
         coord_dtype = getattr(dim_or_data, "dtype", None)
+        # pandas may produce a StringDtype-backed Index, which xarray does
+        # not treat as an allowed extension array dtype for coords. Let
+        # PandasIndex compute a valid numpy coord_dtype in that case so a
+        # subsequent concat against a numpy-string-dtype coord does not fail
+        # in ``np.result_type``. See GH#11317.
+        if isinstance(coord_dtype, pd.StringDtype):
+            coord_dtype = None
         index = PandasIndex(dim_or_data, dim, coord_dtype=coord_dtype)
 
     return dim, index

xarray/tests/test_concat.py

@@ -1491,6 +1491,19 @@ def test_concat_index_not_same_dim() -> None:
         concat([ds1, ds2], dim="x")
 
 
+def test_concat_pandas_index_string_dtype() -> None:
+    # Regression test for GH#11317. When pandas yields a StringDtype-backed
+    # Index (e.g. with ``future.infer_string`` enabled), passing it as the
+    # new concat dim must not break a subsequent concat against a coord with
+    # a numpy string dtype.
+    with pd.option_context("future.infer_string", True):
+        da = DataArray([0], dims=["dim_a"], coords={"dim_a": ["a"]})
+        db = DataArray([0], dims=["dim_b"], coords={"dim_b": ["b"]})
+        db2 = concat([db], pd.Index(["b"], name="dim_a"))
+        result = concat([da, db2], dim="dim_a")
+    assert list(result["dim_a"].values) == ["a", "b"]
+
+
 class TestNewDefaults:
     def test_concat_second_empty_with_scalar_data_var_only_on_first(self) -> None:
         ds1 = Dataset(data_vars={"a": ("y", [0.1]), "b": 0.1}, coords={"x": 0.1})