Erlend mentioned that he'll have ideas for follow-up PRs, so keeping the issue open.

I'll just post my suggestion here instead of opening a PR. It kind of bothers me a bit that PyObject_GetTypeData and PyType_GetTypeDataSize currently always succeeds (and thus never raise an exception), but both are documented to raise an exception on error. I feel the urge to assert the current behaviour so we suddenly don't end up with them returning NULL or -1 without an exception set. What do you think, Petr, is it worth it?

suggestion - diff

diff --git a/Objects/typeobject.c b/Objects/typeobject.c
index 171c76a59a..ad7b9c1ce4 100644
--- a/Objects/typeobject.c
+++ b/Objects/typeobject.c
@@ -4532,7 +4532,12 @@ void *
 PyObject_GetTypeData(PyObject *obj, PyTypeObject *cls)
 {
     assert(PyObject_TypeCheck(obj, cls));
-    return (char *)obj + _align_up(cls->tp_base->tp_basicsize);
+    void *ret = (char *)obj + _align_up(cls->tp_base->tp_basicsize);
+
+    /* This API is documented to raise an exception if it returns NULL.
+     * The current implementation, however, always succeeds. */
+    assert(ret != NULL);
+    return ret;
 }
 
 Py_ssize_t
@@ -4542,6 +4547,10 @@ PyType_GetTypeDataSize(PyTypeObject *cls)
     if (result < 0) {
         return 0;
     }
+
+    /* This API is documented to raise an exception if it returns -1.
+     * The current implementation, however, always succeeds. */
+     assert(result >= 0);
     return result;
 }

Uh oh!

PEP 697 -- Limited C API for Extending Opaque Types #103509

Description

Linked PRs

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions