Make to_numeric default to correct precision (#36149)

Author: Dr-Irv

doc/source/whatsnew/v1.2.0.rst

@@ -245,7 +245,7 @@ Timezones
 
 Numeric
 ^^^^^^^
--
+- Bug in :func:`to_numeric` where float precision was incorrect (:issue:`31364`)
 -
 
 Conversion

pandas/_libs/src/parse_helper.h

@@ -18,7 +18,9 @@ int to_double(char *item, double *p_value, char sci, char decimal,
     char *p_end = NULL;
     int error = 0;
 
-    *p_value = xstrtod(item, &p_end, decimal, sci, '\0', 1, &error, maybe_int);
+    /* Switch to precise xstrtod GH 31364 */
+    *p_value = precise_xstrtod(item, &p_end, decimal, sci, '\0', 1,
+                               &error, maybe_int);
 
     return (error == 0) && (!*p_end);
 }

pandas/tests/tools/test_to_numeric.py

@@ -649,3 +649,61 @@ def test_failure_to_convert_uint64_string_to_NaN():
     ser = Series([32, 64, np.nan])
     result = to_numeric(pd.Series(["32", "64", "uint64"]), errors="coerce")
     tm.assert_series_equal(result, ser)
+
+
+@pytest.mark.parametrize(
+    "strrep",
+    [
+        "243.164",
+        "245.968",
+        "249.585",
+        "259.745",
+        "265.742",
+        "272.567",
+        "279.196",
+        "280.366",
+        "275.034",
+        "271.351",
+        "272.889",
+        "270.627",
+        "280.828",
+        "290.383",
+        "308.153",
+        "319.945",
+        "336.0",
+        "344.09",
+        "351.385",
+        "356.178",
+        "359.82",
+        "361.03",
+        "367.701",
+        "380.812",
+        "387.98",
+        "391.749",
+        "391.171",
+        "385.97",
+        "385.345",
+        "386.121",
+        "390.996",
+        "399.734",
+        "413.073",
+        "421.532",
+        "430.221",
+        "437.092",
+        "439.746",
+        "446.01",
+        "451.191",
+        "460.463",
+        "469.779",
+        "472.025",
+        "479.49",
+        "474.864",
+        "467.54",
+        "471.978",
+    ],
+)
+def test_precision_float_conversion(strrep):
+    # GH 31364
+    result = to_numeric(strrep)
+
+    assert result == float(strrep)