feat(type-system): infer pandas' string dtype

ibis/backends/pandas/client.py

@@ -113,10 +113,15 @@ def from_pandas_tzdtype(value):
 
 
 @dt.dtype.register(CategoricalDtype)
-def from_pandas_categorical(value):
+def from_pandas_categorical(_):
     return dt.Category()
 
 
+@dt.dtype.register(pd.core.arrays.string_.StringDtype)
+def from_pandas_string(_):
+    return dt.String()
+
+
 @dt.infer.register(np.generic)
 def infer_numpy_scalar(value):
     return dt.dtype(value.dtype)
@@ -206,7 +211,7 @@ def infer_pandas_schema(df, schema=None):
     schema = schema if schema is not None else {}
 
     pairs = []
-    for column_name, pandas_dtype in df.dtypes.iteritems():
+    for column_name in df.dtypes.keys():
         if not isinstance(column_name, str):
             raise TypeError(
                 'Column names must be strings to use the pandas backend'

ibis/backends/pandas/tests/test_datatypes.py

@@ -130,6 +130,7 @@ def test_numpy_dtype(numpy_dtype, ibis_dtype):
             dt.Timestamp('US/Eastern'),
         ),
         (CategoricalDtype(), dt.Category()),
+        (pd.Series([], dtype="string").dtype, dt.String()),
     ],
 )
 def test_pandas_dtype(pandas_dtype, ibis_dtype):
@@ -206,6 +207,7 @@ def test_pandas_dtype(pandas_dtype, ibis_dtype):
         (pd.Series([b'1', '2', 3.0]), dt.binary),
         # empty
         (pd.Series([], dtype='object'), dt.binary),
+        (pd.Series([], dtype="string"), dt.string),
     ],
 )
 def test_schema_infer(col_data, schema_type):
@@ -214,3 +216,10 @@ def test_schema_infer(col_data, schema_type):
     inferred = sch.infer(df)
     expected = ibis.schema([('col', schema_type)])
     assert inferred == expected
+
+
+def test_pyarrow_string():
+    pytest.importorskip("pyarrow")
+
+    s = pd.Series([], dtype="string[pyarrow]")
+    assert dt.dtype(s.dtype) == dt.String()