Skip to content

Commit 8ef3460

Browse files
Issue #26906: Resolving special methods of uninitialized type now causes
implicit initialization of the type instead of a fail.
1 parent 1c1130f commit 8ef3460

2 files changed

Lines changed: 22 additions & 5 deletions

File tree

‎Misc/NEWS‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -10,6 +10,9 @@ Release date: TBA
1010
Core and Builtins
1111
-----------------
1212

13+
- Issue #26906: Resolving special methods of uninitialized type now causes
14+
implicit initialization of the type instead of a fail.
15+
1316
- Issue #18287: PyType_Ready() now checks that tp_name is not NULL.
1417
Original patch by Niklas Koep.
1518

‎Objects/typeobject.c‎

Lines changed: 19 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -2869,11 +2869,25 @@ _PyType_Lookup(PyTypeObject *type, PyObject *name)
28692869
/* Look in tp_dict of types in MRO */
28702870
mro = type->tp_mro;
28712871

2872-
/* If mro is NULL, the type is either not yet initialized
2873-
by PyType_Ready(), or already cleared by type_clear().
2874-
Either way the safest thing to do is to return NULL. */
2875-
if (mro == NULL)
2876-
return NULL;
2872+
if (mro == NULL) {
2873+
if ((type->tp_flags & Py_TPFLAGS_READYING) == 0 &&
2874+
PyType_Ready(type) < 0) {
2875+
/* It's not ideal to clear the error condition,
2876+
but this function is documented as not setting
2877+
an exception, and I don't want to change that.
2878+
When PyType_Ready() can't proceed, it won't
2879+
set the "ready" flag, so future attempts to ready
2880+
the same type will call it again -- hopefully
2881+
in a context that propagates the exception out.
2882+
*/
2883+
PyErr_Clear();
2884+
return NULL;
2885+
}
2886+
mro = type->tp_mro;
2887+
if (mro == NULL) {
2888+
return NULL;
2889+
}
2890+
}
28772891

28782892
res = NULL;
28792893
/* keep a strong reference to mro because type->tp_mro can be replaced

0 commit comments

Comments
 (0)