nanobind: fix py::str(accessor) reinterpret bug (wrap in py::object so PyObject_Str runs) across numpy/pandas/udf/replacement paths

Author: evertlammerts

src/duckdb_py/map.cpp

@@ -52,7 +52,7 @@ static py::object FunctionCall(NumpyResultConversion &conversion, const vector<I
 	if (!py::isinstance<PandasDataFrame>(df)) {
 		throw InvalidInputException(
 		    "Expected the UDF to return an object of type 'pandas.DataFrame', found '%s' instead",
-		    py::cast<std::string>(py::str(df.attr("__class__"))));
+		    py::cast<std::string>(py::str(py::object(df.attr("__class__")))));
 	}
 	if (PandasDataFrame::IsPyArrowBacked(df)) {
 		throw InvalidInputException(

src/duckdb_py/numpy/numpy_bind.cpp

@@ -15,11 +15,11 @@ void NumpyBind::Bind(ClientContext &context, py::handle df, vector<PandasColumnB
 	auto df_columns = py::list(df.attr("keys")());
 	auto df_types = py::list();
 	for (auto item : py::cast<py::dict>(df)) {
-		if (py::cast<std::string>(py::str(item.second.attr("dtype").attr("char"))) == "U") {
+		if (py::cast<std::string>(py::str(py::object(item.second.attr("dtype").attr("char")))) == "U") {
 			df_types.attr("append")(py::str("string"));
 			continue;
 		}
-		df_types.attr("append")(py::str(item.second.attr("dtype")));
+		df_types.attr("append")(py::str(py::object(item.second.attr("dtype"))));
 	}
 	auto get_fun = df.attr("__getitem__");
 	if (py::len(df_columns) == 0 || py::len(df_types) == 0 || py::len(df_columns) != py::len(df_types)) {
@@ -53,7 +53,7 @@ void NumpyBind::Bind(ClientContext &context, py::handle df, vector<PandasColumnB
 			}
 			duckdb_col_type = LogicalType::ENUM(enum_entries_vec, size);
 			auto pandas_col = uniq.attr("__getitem__")(1);
-			bind_data.internal_categorical_type = py::cast<std::string>(py::str(pandas_col.attr("dtype")));
+			bind_data.internal_categorical_type = py::cast<std::string>(py::str(py::object(pandas_col.attr("dtype"))));
 			bind_data.pandas_col = std::make_unique<PandasNumpyColumn>(NumpyArray(pandas_col));
 		} else {
 			bind_data.pandas_col = std::make_unique<PandasNumpyColumn>(NumpyArray(column));

src/duckdb_py/pyexpression.cpp

@@ -319,7 +319,7 @@ std::shared_ptr<DuckDBPyExpression> DuckDBPyExpression::StarExpression(py::objec
 std::shared_ptr<DuckDBPyExpression> DuckDBPyExpression::ColumnExpression(const py::args &names) {
 	vector<Identifier> column_names;
 	if (names.size() == 1) {
-		string column_name = py::cast<std::string>(py::str(names[0]));
+		string column_name = py::cast<std::string>(py::str(py::object(names[0])));
 		if (column_name == "*") {
 			return StarExpression();
 		}

src/duckdb_py/python_replacement_scan.cpp

@@ -225,7 +225,7 @@ static unique_ptr<TableRef> TryReplacement(py::dict &dict, const string &name, C
 	if (!result) {
 		std::string location = py::cast<std::string>(current_frame.attr("f_code").attr("co_filename"));
 		location += ":";
-		location += py::cast<std::string>(py::str(current_frame.attr("f_lineno")));
+		location += py::cast<std::string>(py::str(py::object(current_frame.attr("f_lineno"))));
 		ThrowScanFailureError(entry, name, location);
 	}
 	return result;

src/duckdb_py/python_udf.cpp

@@ -400,7 +400,7 @@ static bool NumpyDeprecatesAccessToCore(const py::tuple &numpy_version) {
 	if (numpy_version.empty()) {
 		return false;
 	}
-	if (py::cast<std::string>(py::str(numpy_version[0])) == string("2")) {
+	if (py::cast<std::string>(py::str(py::object(numpy_version[0]))) == string("2")) {
 		//! Starting with numpy version 2.0.0 the use of 'core' is deprecated.
 		return true;
 	}

src/duckdb_py/typing/pytype.cpp

@@ -140,7 +140,7 @@ static bool FromNumpyType(const py::object &type, LogicalType &result) {
 	if (!py::hasattr(obj, "dtype")) {
 		return false;
 	}
-	string type_str = py::cast<std::string>(py::str(obj.attr("dtype")));
+	string type_str = py::cast<std::string>(py::str(py::object(obj.attr("dtype"))));
 	if (type_str == "bool") {
 		result = LogicalType::BOOLEAN;
 	} else if (type_str == "int8") {