coords: retain str dtype (#4759)

* coords: retain str dtype

* fix doctests

* update what's new

* fix multiindex repr

* rename function

* ensure minimum str dtype

* fix EOL spaces

Author: mathause

doc/whats-new.rst

@@ -66,6 +66,9 @@ Bug fixes
   By `Anderson Banihirwe <https://github.com/andersy005>`_
 - Fix a crash in orthogonal indexing on geographic coordinates with ``engine='cfgrib'`` (:issue:`4733` :pull:`4737`).
   By `Alessandro Amici <https://github.com/alexamici>`_
+- Coordinates with dtype ``str`` or ``bytes`` now retain their dtype on many operations,
+  e.g. ``reindex``, ``align``, ``concat``, ``assign``, previously they were cast to an object dtype
+  (:issue:`2658` and :issue:`4543`) by `Mathias Hauser <https://github.com/mathause>`_.
 - Limit number of data rows when printing large datasets. (:issue:`4736`, :pull:`4750`). By `Jimmy Westling <https://github.com/illviljan>`_.
 - Add ``missing_dims`` parameter to transpose (:issue:`4647`, :pull:`4767`). By `Daniel Mesejo <https://github.com/mesejo>`_.
 - Resolve intervals before appending other metadata to labels when plotting (:issue:`4322`, :pull:`4794`).

xarray/core/alignment.py

@@ -19,7 +19,7 @@
 
 from . import dtypes, utils
 from .indexing import get_indexer_nd
-from .utils import is_dict_like, is_full_slice
+from .utils import is_dict_like, is_full_slice, maybe_coerce_to_str
 from .variable import IndexVariable, Variable
 
 if TYPE_CHECKING:
@@ -278,10 +278,12 @@ def align(
         return (obj.copy(deep=copy),)
 
     all_indexes = defaultdict(list)
+    all_coords = defaultdict(list)
     unlabeled_dim_sizes = defaultdict(set)
     for obj in objects:
         for dim in obj.dims:
             if dim not in exclude:
+                all_coords[dim].append(obj.coords[dim])
                 try:
                     index = obj.indexes[dim]
                 except KeyError:
@@ -306,7 +308,7 @@ def align(
                 any(not index.equals(other) for other in matching_indexes)
                 or dim in unlabeled_dim_sizes
             ):
-                joined_indexes[dim] = index
+                joined_indexes[dim] = indexes[dim]
         else:
             if (
                 any(
@@ -318,9 +320,11 @@ def align(
                 if join == "exact":
                     raise ValueError(f"indexes along dimension {dim!r} are not equal")
                 index = joiner(matching_indexes)
+                # make sure str coords are not cast to object
+                index = maybe_coerce_to_str(index, all_coords[dim])
                 joined_indexes[dim] = index
             else:
-                index = matching_indexes[0]
+                index = all_coords[dim][0]
 
         if dim in unlabeled_dim_sizes:
             unlabeled_sizes = unlabeled_dim_sizes[dim]
@@ -583,7 +587,7 @@ def reindex_variables(
             args: tuple = (var.attrs, var.encoding)
         else:
             args = ()
-        reindexed[dim] = IndexVariable((dim,), target, *args)
+        reindexed[dim] = IndexVariable((dim,), indexers[dim], *args)
 
     for dim in sizes:
         if dim not in indexes and dim in indexers:

xarray/core/concat.py

@@ -187,7 +187,7 @@ def concat(
     array([[0, 1, 2],
            [3, 4, 5]])
     Coordinates:
-      * x        (x) object 'a' 'b'
+      * x        (x) <U1 'a' 'b'
       * y        (y) int64 10 20 30
 
     >>> xr.concat([da.isel(x=0), da.isel(x=1)], "new_dim")
@@ -503,7 +503,7 @@ def ensure_common_dims(vars):
     for k in datasets[0].variables:
         if k in concat_over:
             try:
-                vars = ensure_common_dims([ds.variables[k] for ds in datasets])
+                vars = ensure_common_dims([ds[k].variable for ds in datasets])
             except KeyError:
                 raise ValueError("%r is not present in all datasets." % k)
             combined = concat_vars(vars, dim, positions)

xarray/core/dataarray.py

@@ -1325,8 +1325,8 @@ def broadcast_like(
                [ 2.2408932 ,  1.86755799, -0.97727788],
                [        nan,         nan,         nan]])
         Coordinates:
-          * x        (x) object 'a' 'b' 'c'
-          * y        (y) object 'a' 'b' 'c'
+          * x        (x) <U1 'a' 'b' 'c'
+          * y        (y) <U1 'a' 'b' 'c'
         """
         if exclude is None:
             exclude = set()

xarray/core/dataset.py

@@ -2565,7 +2565,7 @@ def reindex(
         <xarray.Dataset>
         Dimensions:      (station: 4)
         Coordinates:
-          * station      (station) object 'boston' 'austin' 'seattle' 'lincoln'
+          * station      (station) <U7 'boston' 'austin' 'seattle' 'lincoln'
         Data variables:
             temperature  (station) float64 10.98 nan 12.06 nan
             pressure     (station) float64 211.8 nan 218.8 nan
@@ -2576,7 +2576,7 @@ def reindex(
         <xarray.Dataset>
         Dimensions:      (station: 4)
         Coordinates:
-          * station      (station) object 'boston' 'austin' 'seattle' 'lincoln'
+          * station      (station) <U7 'boston' 'austin' 'seattle' 'lincoln'
         Data variables:
             temperature  (station) float64 10.98 0.0 12.06 0.0
             pressure     (station) float64 211.8 0.0 218.8 0.0
@@ -2589,7 +2589,7 @@ def reindex(
         <xarray.Dataset>
         Dimensions:      (station: 4)
         Coordinates:
-          * station      (station) object 'boston' 'austin' 'seattle' 'lincoln'
+          * station      (station) <U7 'boston' 'austin' 'seattle' 'lincoln'
         Data variables:
             temperature  (station) float64 10.98 0.0 12.06 0.0
             pressure     (station) float64 211.8 100.0 218.8 100.0

xarray/core/merge.py

@@ -930,9 +930,11 @@ def dataset_update_method(
                 if coord_names:
                     other[key] = value.drop_vars(coord_names)
 
+    # use ds.coords and not ds.indexes, else str coords are cast to object
+    indexes = {key: dataset.coords[key] for key in dataset.indexes.keys()}
     return merge_core(
         [dataset, other],
         priority_arg=1,
-        indexes=dataset.indexes,
+        indexes=indexes,
         combine_attrs="override",
     )

xarray/core/utils.py

@@ -31,6 +31,8 @@
 import numpy as np
 import pandas as pd
 
+from . import dtypes
+
 K = TypeVar("K")
 V = TypeVar("V")
 T = TypeVar("T")
@@ -76,6 +78,23 @@ def maybe_cast_to_coords_dtype(label, coords_dtype):
     return label
 
 
+def maybe_coerce_to_str(index, original_coords):
+    """maybe coerce a pandas Index back to a nunpy array of type str
+
+    pd.Index uses object-dtype to store str - try to avoid this for coords
+    """
+
+    try:
+        result_type = dtypes.result_type(*original_coords)
+    except TypeError:
+        pass
+    else:
+        if result_type.kind in "SU":
+            index = np.asarray(index, dtype=result_type.type)
+
+    return index
+
+
 def safe_cast_to_index(array: Any) -> pd.Index:
     """Given an array, safely cast it to a pandas.Index.
 

xarray/core/variable.py

@@ -48,6 +48,7 @@
     ensure_us_time_resolution,
     infix_dims,
     is_duck_array,
+    maybe_coerce_to_str,
 )
 
 NON_NUMPY_SUPPORTED_ARRAY_TYPES = (
@@ -2523,6 +2524,9 @@ def concat(cls, variables, dim="concat_dim", positions=None, shortcut=False):
                 indices = nputils.inverse_permutation(np.concatenate(positions))
                 data = data.take(indices)
 
+        # keep as str if possible as pandas.Index uses object (converts to numpy array)
+        data = maybe_coerce_to_str(data, variables)
+
         attrs = dict(first_var.attrs)
         if not shortcut:
             for var in variables:

xarray/tests/test_concat.py

@@ -376,6 +376,30 @@ def test_concat_fill_value(self, fill_value):
         actual = concat(datasets, dim="t", fill_value=fill_value)
         assert_identical(actual, expected)
 
+    @pytest.mark.parametrize("dtype", [str, bytes])
+    @pytest.mark.parametrize("dim", ["x1", "x2"])
+    def test_concat_str_dtype(self, dtype, dim):
+
+        data = np.arange(4).reshape([2, 2])
+
+        da1 = Dataset(
+            {
+                "data": (["x1", "x2"], data),
+                "x1": [0, 1],
+                "x2": np.array(["a", "b"], dtype=dtype),
+            }
+        )
+        da2 = Dataset(
+            {
+                "data": (["x1", "x2"], data),
+                "x1": np.array([1, 2]),
+                "x2": np.array(["c", "d"], dtype=dtype),
+            }
+        )
+        actual = concat([da1, da2], dim=dim)
+
+        assert np.issubdtype(actual.x2.dtype, dtype)
+
 
 class TestConcatDataArray:
     def test_concat(self):
@@ -525,6 +549,26 @@ def test_concat_combine_attrs_kwarg(self):
             actual = concat([da1, da2], dim="x", combine_attrs=combine_attrs)
             assert_identical(actual, expected[combine_attrs])
 
+    @pytest.mark.parametrize("dtype", [str, bytes])
+    @pytest.mark.parametrize("dim", ["x1", "x2"])
+    def test_concat_str_dtype(self, dtype, dim):
+
+        data = np.arange(4).reshape([2, 2])
+
+        da1 = DataArray(
+            data=data,
+            dims=["x1", "x2"],
+            coords={"x1": [0, 1], "x2": np.array(["a", "b"], dtype=dtype)},
+        )
+        da2 = DataArray(
+            data=data,
+            dims=["x1", "x2"],
+            coords={"x1": np.array([1, 2]), "x2": np.array(["c", "d"], dtype=dtype)},
+        )
+        actual = concat([da1, da2], dim=dim)
+
+        assert np.issubdtype(actual.x2.dtype, dtype)
+
 
 @pytest.mark.parametrize("attr1", ({"a": {"meta": [10, 20, 30]}}, {"a": [1, 2, 3]}, {}))
 @pytest.mark.parametrize("attr2", ({"a": [1, 2, 3]}, {}))

xarray/tests/test_dataarray.py

@@ -1568,6 +1568,19 @@ def test_reindex_fill_value(self, fill_value):
         )
         assert_identical(expected, actual)
 
+    @pytest.mark.parametrize("dtype", [str, bytes])
+    def test_reindex_str_dtype(self, dtype):
+
+        data = DataArray(
+            [1, 2], dims="x", coords={"x": np.array(["a", "b"], dtype=dtype)}
+        )
+
+        actual = data.reindex(x=data.x)
+        expected = data
+
+        assert_identical(expected, actual)
+        assert actual.dtype == expected.dtype
+
     def test_rename(self):
         renamed = self.dv.rename("bar")
         assert_identical(renamed.to_dataset(), self.ds.rename({"foo": "bar"}))
@@ -3435,6 +3448,26 @@ def test_align_without_indexes_errors(self):
                 DataArray([1, 2], coords=[("x", [0, 1])]),
             )
 
+    def test_align_str_dtype(self):
+
+        a = DataArray([0, 1], dims=["x"], coords={"x": ["a", "b"]})
+        b = DataArray([1, 2], dims=["x"], coords={"x": ["b", "c"]})
+
+        expected_a = DataArray(
+            [0, 1, np.NaN], dims=["x"], coords={"x": ["a", "b", "c"]}
+        )
+        expected_b = DataArray(
+            [np.NaN, 1, 2], dims=["x"], coords={"x": ["a", "b", "c"]}
+        )
+
+        actual_a, actual_b = xr.align(a, b, join="outer")
+
+        assert_identical(expected_a, actual_a)
+        assert expected_a.x.dtype == actual_a.x.dtype
+
+        assert_identical(expected_b, actual_b)
+        assert expected_b.x.dtype == actual_b.x.dtype
+
     def test_broadcast_arrays(self):
         x = DataArray([1, 2], coords=[("a", [-1, -2])], name="x")
         y = DataArray([1, 2], coords=[("b", [3, 4])], name="y")