BUG: read_csv not converting to float for python engine with decimal sep, usecols and parse_dates (#38334)

Author: phofl

doc/source/whatsnew/v1.3.0.rst

@@ -264,6 +264,7 @@ I/O
 - Bug in :func:`read_csv` interpreting ``NA`` value as comment, when ``NA`` does contain the comment string fixed for ``engine="python"`` (:issue:`34002`)
 - Bug in :func:`read_csv` raising ``IndexError`` with multiple header columns and ``index_col`` specified when file has no data rows (:issue:`38292`)
 - Bug in :func:`read_csv` not accepting ``usecols`` with different length than ``names`` for ``engine="python"`` (:issue:`16469`)
+- Bug in :meth:`read_csv` returning object dtype when ``delimiter=","`` with ``usecols`` and ``parse_dates`` specified for ``engine="python"`` (:issue:`35873`)
 - Bug in :func:`read_csv` raising ``TypeError`` when ``names`` and ``parse_dates`` is specified for ``engine="c"`` (:issue:`33699`)
 - Bug in :func:`read_clipboard`, :func:`DataFrame.to_clipboard` not working in WSL (:issue:`38527`)
 - Allow custom error values for parse_dates argument of :func:`read_sql`, :func:`read_sql_query` and :func:`read_sql_table` (:issue:`35185`)

pandas/io/parsers.py

@@ -2293,7 +2293,7 @@ def __init__(self, f: Union[FilePathOrBuffer, List], **kwds):
 
         # Get columns in two steps: infer from data, then
         # infer column indices from self.usecols if it is specified.
-        self._col_indices = None
+        self._col_indices: Optional[List[int]] = None
         try:
             (
                 self.columns,
@@ -2335,6 +2335,9 @@ def __init__(self, f: Union[FilePathOrBuffer, List], **kwds):
             if self.index_names is None:
                 self.index_names = index_names
 
+        if self._col_indices is None:
+            self._col_indices = list(range(len(self.columns)))
+
         self._validate_parse_dates_presence(self.columns)
         if self.parse_dates:
             self._no_thousands_columns = self._set_no_thousands_columns()
@@ -2364,7 +2367,9 @@ def _set(x):
             if is_integer(x):
                 noconvert_columns.add(x)
             else:
-                noconvert_columns.add(self.columns.index(x))
+                assert self._col_indices is not None
+                col_indices = self._col_indices
+                noconvert_columns.add(col_indices[self.columns.index(x)])
 
         if isinstance(self.parse_dates, list):
             for val in self.parse_dates:
@@ -2706,7 +2711,6 @@ def _infer_columns(self):
                     # overwritten.
                     self._handle_usecols(columns, names)
                 else:
-                    self._col_indices = None
                     num_original_columns = len(names)
                 columns = [names]
             else:
@@ -2788,7 +2792,7 @@ def _handle_usecols(self, columns, usecols_key):
                 [n for i, n in enumerate(column) if i in col_indices]
                 for column in columns
             ]
-            self._col_indices = col_indices
+            self._col_indices = sorted(col_indices)
         return columns
 
     def _buffered_line(self):
@@ -3186,25 +3190,21 @@ def _rows_to_cols(self, content):
         zipped_content = list(lib.to_object_array(content, min_width=col_len).T)
 
         if self.usecols:
+            assert self._col_indices is not None
+            col_indices = self._col_indices
+
             if self._implicit_index:
                 zipped_content = [
                     a
                     for i, a in enumerate(zipped_content)
                     if (
                         i < len(self.index_col)
-                        # pandas\io\parsers.py:3159: error: Unsupported right
-                        # operand type for in ("Optional[Any]")  [operator]
-                        or i - len(self.index_col)  # type: ignore[operator]
-                        in self._col_indices
+                        or i - len(self.index_col) in col_indices
                     )
                 ]
             else:
                 zipped_content = [
-                    # pandas\io\parsers.py:3164: error: Unsupported right
-                    # operand type for in ("Optional[Any]")  [operator]
-                    a
-                    for i, a in enumerate(zipped_content)
-                    if i in self._col_indices  # type: ignore[operator]
+                    a for i, a in enumerate(zipped_content) if i in col_indices
                 ]
         return zipped_content
 

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

@@ -10,7 +10,7 @@
 from pandas.errors import ParserWarning
 
 import pandas as pd
-from pandas import DataFrame
+from pandas import DataFrame, Timestamp
 import pandas._testing as tm
 
 
@@ -165,3 +165,19 @@ def test_boolean_dtype(all_parsers):
     )
 
     tm.assert_frame_equal(result, expected)
+
+
+def test_delimiter_with_usecols_and_parse_dates(all_parsers):
+    # GH#35873
+    result = all_parsers.read_csv(
+        StringIO('"dump","-9,1","-9,1",20101010'),
+        engine="python",
+        names=["col", "col1", "col2", "col3"],
+        usecols=["col1", "col2", "col3"],
+        parse_dates=["col3"],
+        decimal=",",
+    )
+    expected = DataFrame(
+        {"col1": [-9.1], "col2": [-9.1], "col3": [Timestamp("2010-10-10")]}
+    )
+    tm.assert_frame_equal(result, expected)