gh-131711: Preventing the use of a null pointer in set_tp_mro

[smurav]

In some cases, a null pointer may be used in function set_tp_mro.

• Issue: Using a null pointer in set_tp_mro #131711

[vstinner]

LGTM

[vstinner]

I can reproduce a crash using this local change:

diff --git a/Objects/typeobject.c b/Objects/typeobject.c
index fc4950ef30a..6e880c5235a 100644
--- a/Objects/typeobject.c
+++ b/Objects/typeobject.c
@@ -1658,7 +1658,11 @@ mro_hierarchy(PyTypeObject *type, PyObject *temp)
         tuple = PyTuple_Pack(3, type, new_mro, old_mro);
     }
     else {
+#if 0
         tuple = PyTuple_Pack(2, type, new_mro);
+#else
+        tuple = NULL;
+#endif
     }
 
     if (tuple != NULL) {

if I run the test:

$ ./python -m test -v test_descr -m test_incomplete_set_bases_on_self
(...)
test_incomplete_set_bases_on_self (test.test_descr.MroTest.test_incomplete_set_bases_on_self)
type_set_bases must be aware that type->tp_mro can be NULL. ... Fatal Python error: Segmentation fault

Current thread 0x00007f549a70d740 (most recent call first):
  File "/home/vstinner/python/main/Lib/test/test_descr.py", line 5778 in mro
  File "/home/vstinner/python/main/Lib/test/test_descr.py", line 5748 in __new__
  File "/home/vstinner/python/main/Lib/test/test_descr.py", line 5782 in test_incomplete_set_bases_on_self
(...)

[vstinner]

Merged, thanks for the fix.