Fix/na_values_GH59303

doc/source/whatsnew/v3.0.0.rst

@@ -706,6 +706,7 @@ Other
 - Bug in :meth:`Series.dt` methods in :class:`ArrowDtype` that were returning incorrect values. (:issue:`57355`)
 - Bug in :meth:`Series.rank` that doesn't preserve missing values for nullable integers when ``na_option='keep'``. (:issue:`56976`)
 - Bug in :meth:`Series.replace` and :meth:`DataFrame.replace` inconsistently replacing matching instances when ``regex=True`` and missing values are present. (:issue:`56599`)
+- Bug in :meth:`_clean_na_values` in :class:`TextFileReader` that was not properly handling ``na_values`` when it is a list of strings. (:issue:`59303`)
 - Bug in Dataframe Interchange Protocol implementation was returning incorrect results for data buffers' associated dtype, for string and datetime columns (:issue:`54781`)
 - Bug in ``Series.list`` methods not preserving the original :class:`Index`. (:issue:`58425`)
 

pandas/io/parsers/readers.py

@@ -1648,7 +1648,7 @@ def _clean_na_values(na_values, keep_default_na: bool = True, floatify: bool = T
             if keep_default_na:
                 v = set(v) | STR_NA_VALUES
 
-            na_values[k] = v
+            na_values[k] = _stringify_na_values(v, floatify)
         na_fvalues = {k: _floatify_na_values(v) for k, v in na_values.items()}
     else:
         if not is_list_like(na_values):

pandas/tests/io/parser/test_na_values.py

@@ -812,3 +812,23 @@ def test_bool_and_nan_to_float(all_parsers):
     result = parser.read_csv(StringIO(data), dtype="float")
     expected = DataFrame.from_dict({"0": [np.nan, 1.0, 0.0]})
     tm.assert_frame_equal(result, expected)
+
+
+@xfail_pyarrow
+def test_na_values_dict_without_dtype(all_parsers):
+    # GH#59303
+    parser = all_parsers
+    data = """A
+-99
+-99
+-99.0
+-99.0"""
+    # this would FAIL BEFORE this fix
+    result_1 = parser.read_csv(StringIO(data), na_values={"A": [-99.0, -99]})
+    expected_1 = DataFrame.from_dict({"A": [np.nan, np.nan, np.nan, np.nan]})
+    tm.assert_frame_equal(result_1, expected_1)
+
+    # this would PASS even BEFORE this fix
+    result_2 = parser.read_csv(StringIO(data), na_values={"A": [-99, -99.0]})
+    expected_2 = DataFrame.from_dict({"A": [np.nan, np.nan, np.nan, np.nan]})
+    tm.assert_frame_equal(result_2, expected_2)