gh-126024: fix UBSan failure in `unicodeobject.c:find_first_nonascii` (GH-127566)

diff --git a/Objects/unicodeobject.c b/Objects/unicodeobject.c
index 463da06445984b5dd9d4fc6acfc29869c2d9711f..33c4747bbef4883def8e22e30cd6c1f0fa5c4408 100644 (file)
--- a/Objects/unicodeobject.c
+++ b/Objects/unicodeobject.c
@@ -5083,12 +5083,9 @@ find_first_nonascii(const unsigned char *start, const unsigned char *end)
         const unsigned char *p2 = _Py_ALIGN_UP(p, SIZEOF_SIZE_T);
 #if PY_LITTLE_ENDIAN && HAVE_CTZ
         if (p < p2) {
-#if defined(_M_AMD64) || defined(_M_IX86) || defined(__x86_64__) || defined(__i386__)
-            // x86 and amd64 are little endian and can load unaligned memory.
-            size_t u = *(const size_t*)p & ASCII_CHAR_MASK;
-#else
-            size_t u = load_unaligned(p, p2 - p) & ASCII_CHAR_MASK;
-#endif
+            size_t u;
+            memcpy(&u, p, sizeof(size_t));
+            u &= ASCII_CHAR_MASK;
             if (u) {
                 return (ctz(u) - 7) / 8;
             }