fix: Avoid name collisions and panicking in object conversion

crates/polars-python/src/conversion/any_value.rs

@@ -471,15 +471,15 @@ pub(crate) fn py_object_to_any_value<'py>(
             Ok(get_struct)
         } else {
             let ob_type = ob.get_type();
-            let type_name = ob_type.qualname()?.to_string();
+            let type_name = ob_type.fully_qualified_name()?.to_string();
             match type_name.as_str() {
                 // Can't use pyo3::types::PyDateTime with abi3-py37 feature,
                 // so need this workaround instead of `isinstance(ob, datetime)`.
-                "date" => Ok(get_date as InitFn),
-                "time" => Ok(get_time as InitFn),
-                "datetime" => Ok(get_datetime as InitFn),
-                "timedelta" => Ok(get_timedelta as InitFn),
-                "Decimal" => Ok(get_decimal as InitFn),
+                "datetime.date" => Ok(get_date as InitFn),
+                "datetime.time" => Ok(get_time as InitFn),
+                "datetime.datetime" => Ok(get_datetime as InitFn),
+                "datetime.timedelta" => Ok(get_timedelta as InitFn),
+                "decimal.Decimal" => Ok(get_decimal as InitFn),
                 "range" => Ok(get_list as InitFn),
                 _ => {
                     // Support NumPy scalars.

py-polars/tests/unit/datatypes/test_object.py

@@ -239,3 +239,16 @@ def test_object_estimated_size() -> None:
 
     # is a huge underestimation
     assert df.estimated_size() == 9
+
+
+def test_object_polars_dtypes_20572() -> None:
+    df = pl.DataFrame(
+        {
+            "a": pl.Date(),
+            "b": pl.Decimal(5, 1),
+            "c": pl.Int64(),
+            "d": pl.Object(),
+            "e": pl.String(),
+        }
+    )
+    assert all(dt == pl.Object() for dt in df.schema.dtypes())