gh-148390: fix undefined behavior of memoryview(...).cast("?")

Author: picnixz

Lib/test/test_memoryview.py

@@ -620,9 +620,21 @@ def check_equal(view, is_equal):
             m = memoryview(a.tobytes()).cast('n')
             check_equal(m, True)
 
-        # Test '?' format
+        # Test '?' format (keep all the checks below for UBSan)
         m = memoryview(b'\0\1\2').cast('?')
         check_equal(m, True)
+        # m1a and m1b are equivalent to [False, True, False]
+        m1a = memoryview(b'\0\2\0').cast('?')
+        self.assertEqual(m1a.tolist(), [False, True, False])
+        m1b = memoryview(b'\0\4\0').cast('?')
+        self.assertEqual(m1b.tolist(), [False, True, False])
+        self.assertEqual(m1a, m1b)
+        # m1a and m1b are equivalent to [True, True, True]
+        m2a = memoryview(b'\1\3\5').cast('?')
+        self.assertEqual(m2a.tolist(), [True, True, True])
+        m2b = memoryview(b'\2\4\6').cast('?')
+        self.assertEqual(m2b.tolist(), [True, True, True])
+        self.assertEqual(m2a, m2b)
 
         # Test float formats
         for float_format in 'fd':

Misc/NEWS.d/next/Core_and_Builtins/2026-04-12-17-27-28.gh-issue-148390.MAhw7F.rst

@@ -0,0 +1,4 @@
+Fix an undefined behavior in :class:`memoryview` when using the native
+boolean format (``?``) in :meth:`~memoryview.cast`. Previously, calling
+``memoryview(b).cast("?").tolist()`` incorrectly returned ``[False]``
+instead of ``[True]`` for any even byte *b*. Patch by Bénédikt Tran.

Objects/memoryobject.c

@@ -1676,6 +1676,9 @@ fix_error_int(const char *fmt)
     return -1;
 }
 
+// Return 0 if PTR represents "false", and 1 otherwise.
+#define UNPACK_TO_BOOL(PTR) (memcmp((PTR), &(_Bool){0}, sizeof(_Bool)) != 0)
+
 /* Accept integer objects or objects with an __index__() method. */
 static long
 pylong_as_ld(PyObject *item)
@@ -1811,7 +1814,7 @@ unpack_single(PyMemoryViewObject *self, const char *ptr, const char *fmt)
     case 'l': UNPACK_SINGLE(ld, ptr, long); goto convert_ld;
 
     /* boolean */
-    case '?': UNPACK_SINGLE(ld, ptr, _Bool); goto convert_bool;
+    case '?': ld = UNPACK_TO_BOOL(ptr); goto convert_bool;
 
     /* unsigned integers */
     case 'H': UNPACK_SINGLE(lu, ptr, unsigned short); goto convert_lu;
@@ -3029,7 +3032,7 @@ unpack_cmp(const char *p, const char *q, char fmt,
     case 'l': CMP_SINGLE(p, q, long); return equal;
 
     /* boolean */
-    case '?': CMP_SINGLE(p, q, _Bool); return equal;
+    case '?': return UNPACK_TO_BOOL(p) == UNPACK_TO_BOOL(q);
 
     /* unsigned integers */
     case 'H': CMP_SINGLE(p, q, unsigned short); return equal;