BUG: Avoid RangeIndex conversion in read_csv if dtype is specified (#59316)

* BUG: Avoid RangeIndex conversion in read_csv if dtype is specified

* Undo change

* Typing

Author: mroeschke

pandas/io/parsers/base_parser.py

@@ -464,7 +464,11 @@ def _agg_index(self, index, try_parse_dates: bool = True) -> Index:
         arrays = []
         converters = self._clean_mapping(self.converters)
 
-        for i, arr in enumerate(index):
+        if self.index_names is not None:
+            names: Iterable = self.index_names
+        else:
+            names = itertools.cycle([None])
+        for i, (arr, name) in enumerate(zip(index, names)):
             if try_parse_dates and self._should_parse_dates(i):
                 arr = self._date_conv(
                     arr,
@@ -504,12 +508,17 @@ def _agg_index(self, index, try_parse_dates: bool = True) -> Index:
             arr, _ = self._infer_types(
                 arr, col_na_values | col_na_fvalues, cast_type is None, try_num_bool
             )
-            arrays.append(arr)
-
-        names = self.index_names
-        index = ensure_index_from_sequences(arrays, names)
+            if cast_type is not None:
+                # Don't perform RangeIndex inference
+                idx = Index(arr, name=name, dtype=cast_type)
+            else:
+                idx = ensure_index_from_sequences([arr], [name])
+            arrays.append(idx)
 
-        return index
+        if len(arrays) == 1:
+            return arrays[0]
+        else:
+            return MultiIndex.from_arrays(arrays)
 
     @final
     def _convert_to_ndarrays(
@@ -1084,12 +1093,11 @@ def _get_empty_meta(self, columns, dtype: DtypeArg | None = None):
         dtype_dict: defaultdict[Hashable, Any]
         if not is_dict_like(dtype):
             # if dtype == None, default will be object.
-            default_dtype = dtype or object
-            dtype_dict = defaultdict(lambda: default_dtype)
+            dtype_dict = defaultdict(lambda: dtype)
         else:
             dtype = cast(dict, dtype)
             dtype_dict = defaultdict(
-                lambda: object,
+                lambda: None,
                 {columns[k] if is_integer(k) else k: v for k, v in dtype.items()},
             )
 
@@ -1106,8 +1114,14 @@ def _get_empty_meta(self, columns, dtype: DtypeArg | None = None):
         if (index_col is None or index_col is False) or index_names is None:
             index = default_index(0)
         else:
-            data = [Series([], dtype=dtype_dict[name]) for name in index_names]
-            index = ensure_index_from_sequences(data, names=index_names)
+            # TODO: We could return default_index(0) if dtype_dict[name] is None
+            data = [
+                Index([], name=name, dtype=dtype_dict[name]) for name in index_names
+            ]
+            if len(data) == 1:
+                index = data[0]
+            else:
+                index = MultiIndex.from_arrays(data)
             index_col.sort()
 
             for i, n in enumerate(index_col):

pandas/tests/io/parser/dtypes/test_dtypes_basic.py

@@ -22,6 +22,8 @@
     "ignore:Passing a BlockManager to DataFrame:DeprecationWarning"
 )
 
+xfail_pyarrow = pytest.mark.usefixtures("pyarrow_xfail")
+
 
 @pytest.mark.parametrize("dtype", [str, object])
 @pytest.mark.parametrize("check_orig", [True, False])
@@ -594,6 +596,7 @@ def test_string_inference_object_dtype(all_parsers, dtype, using_infer_string):
     tm.assert_frame_equal(result, expected)
 
 
+@xfail_pyarrow
 def test_accurate_parsing_of_large_integers(all_parsers):
     # GH#52505
     data = """SYMBOL,MOMENT,ID,ID_DEAL
@@ -604,7 +607,7 @@ def test_accurate_parsing_of_large_integers(all_parsers):
 AMZN,20230301181139587,2023552585717889759,2023552585717263360
 MSFT,20230301181139587,2023552585717889863,2023552585717263361
 NVDA,20230301181139587,2023552585717889827,2023552585717263361"""
-    orders = pd.read_csv(StringIO(data), dtype={"ID_DEAL": pd.Int64Dtype()})
+    orders = all_parsers.read_csv(StringIO(data), dtype={"ID_DEAL": pd.Int64Dtype()})
     assert len(orders.loc[orders["ID_DEAL"] == 2023552585717263358, "ID_DEAL"]) == 1
     assert len(orders.loc[orders["ID_DEAL"] == 2023552585717263359, "ID_DEAL"]) == 1
     assert len(orders.loc[orders["ID_DEAL"] == 2023552585717263360, "ID_DEAL"]) == 2
@@ -626,3 +629,16 @@ def test_dtypes_with_usecols(all_parsers):
         values = ["1", "4"]
     expected = DataFrame({"a": pd.Series(values, dtype=object), "c": [3, 6]})
     tm.assert_frame_equal(result, expected)
+
+
+def test_index_col_with_dtype_no_rangeindex(all_parsers):
+    data = StringIO("345.5,519.5,0\n519.5,726.5,1")
+    result = all_parsers.read_csv(
+        data,
+        header=None,
+        names=["start", "stop", "bin_id"],
+        dtype={"start": np.float32, "stop": np.float32, "bin_id": np.uint32},
+        index_col="bin_id",
+    ).index
+    expected = pd.Index([0, 1], dtype=np.uint32, name="bin_id")
+    tm.assert_index_equal(result, expected)