From 6dafd9a01d1b230fbdbd6748d74f868506371282 Mon Sep 17 00:00:00 2001
From: Inada Naoki <songofacandy@gmail.com>
Date: Sat, 30 Nov 2024 09:06:55 +0900
Subject: [PATCH 1/4] fix assertion error on AIX

---
 Objects/unicodeobject.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/Objects/unicodeobject.c b/Objects/unicodeobject.c
index ab4f07ed054385..85cfc202ca002f 100644
--- a/Objects/unicodeobject.c
+++ b/Objects/unicodeobject.c
@@ -5022,13 +5022,13 @@ ctz(size_t v)
 static size_t
 load_unaligned(const unsigned char *p, size_t size)
 {
-    assert(size <= SIZEOF_SIZE_T);
     union {
         size_t s;
         unsigned char b[SIZEOF_SIZE_T];
     } u;
     u.s = 0;
     switch (size) {
+    default:
     case 8:
         u.b[7] = p[7];
         _Py_FALLTHROUGH;
@@ -5055,8 +5055,6 @@ load_unaligned(const unsigned char *p, size_t size)
         break;
     case 0:
         break;
-    default:
-        Py_UNREACHABLE();
     }
     return u.s;
 }

From 2faf883b3ca2da93525a9ee06c56da3ea36de525 Mon Sep 17 00:00:00 2001
From: Inada Naoki <songofacandy@gmail.com>
Date: Sat, 30 Nov 2024 12:18:20 +0900
Subject: [PATCH 2/4] fix 32bit build

---
 Objects/unicodeobject.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/Objects/unicodeobject.c b/Objects/unicodeobject.c
index 85cfc202ca002f..184c61d94d462b 100644
--- a/Objects/unicodeobject.c
+++ b/Objects/unicodeobject.c
@@ -5029,6 +5029,7 @@ load_unaligned(const unsigned char *p, size_t size)
     u.s = 0;
     switch (size) {
     default:
+#if SIZEOF_SIZE_T == 8
     case 8:
         u.b[7] = p[7];
         _Py_FALLTHROUGH;
@@ -5041,6 +5042,7 @@ load_unaligned(const unsigned char *p, size_t size)
     case 5:
         u.b[4] = p[4];
         _Py_FALLTHROUGH;
+#endif
     case 4:
         u.b[3] = p[3];
         _Py_FALLTHROUGH;

From ba3e008d0ddabcdfe5606cb0478fc23073729e44 Mon Sep 17 00:00:00 2001
From: Inada Naoki <songofacandy@gmail.com>
Date: Sat, 30 Nov 2024 13:37:24 +0900
Subject: [PATCH 3/4] fix for big endian

---
 Objects/unicodeobject.c | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/Objects/unicodeobject.c b/Objects/unicodeobject.c
index 184c61d94d462b..3de1d1bf5e874a 100644
--- a/Objects/unicodeobject.c
+++ b/Objects/unicodeobject.c
@@ -5016,9 +5016,8 @@ ctz(size_t v)
 }
 #endif
 
-#if HAVE_CTZ
-// load p[0]..p[size-1] as a little-endian size_t
-// without unaligned access nor read ahead.
+#if HAVE_CTZ && PY_LITTLE_ENDIAN
+// load p[0]..p[size-1] as a size_t without unaligned access nor read ahead.
 static size_t
 load_unaligned(const unsigned char *p, size_t size)
 {
@@ -5027,6 +5026,9 @@ load_unaligned(const unsigned char *p, size_t size)
         unsigned char b[SIZEOF_SIZE_T];
     } u;
     u.s = 0;
+    // This switch statement assumes little endian because:
+    // * union is faster than bitwise or and shift.
+    // * big endian machine is rare and hard to maintain.
     switch (size) {
     default:
 #if SIZEOF_SIZE_T == 8
@@ -5078,7 +5080,7 @@ find_first_nonascii(const unsigned char *start, const unsigned char *end)
     if (end - start >= SIZEOF_SIZE_T) {
         const unsigned char *p2 = _Py_ALIGN_UP(p, SIZEOF_SIZE_T);
         if (p < p2) {
-#if HAVE_CTZ
+#if PY_LITTLE_ENDIAN && HAVE_CTZ
 #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;
@@ -5113,7 +5115,7 @@ find_first_nonascii(const unsigned char *start, const unsigned char *end)
             p += SIZEOF_SIZE_T;
         }
     }
-#if HAVE_CTZ
+#if PY_LITTLE_ENDIAN && HAVE_CTZ
     // we can not use *(const size_t*)p to avoid buffer overrun.
     size_t u = load_unaligned(p, end - p) & ASCII_CHAR_MASK;
     if (u) {

From aa4a8397e84296e8d1ab88ac3213a44212b58fef Mon Sep 17 00:00:00 2001
From: Inada Naoki <songofacandy@gmail.com>
Date: Sat, 30 Nov 2024 15:48:03 +0900
Subject: [PATCH 4/4] fixup

---
 Objects/unicodeobject.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/Objects/unicodeobject.c b/Objects/unicodeobject.c
index 3de1d1bf5e874a..33fa21d4c7d1bf 100644
--- a/Objects/unicodeobject.c
+++ b/Objects/unicodeobject.c
@@ -5014,6 +5014,8 @@ ctz(size_t v)
 #endif /* SIZEOF_SIZE_T */
     return pos;
 }
+#else
+#define HAVE_CTZ 0
 #endif
 
 #if HAVE_CTZ && PY_LITTLE_ENDIAN
@@ -5079,8 +5081,8 @@ find_first_nonascii(const unsigned char *start, const unsigned char *end)
 
     if (end - start >= SIZEOF_SIZE_T) {
         const unsigned char *p2 = _Py_ALIGN_UP(p, SIZEOF_SIZE_T);
-        if (p < p2) {
 #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;
@@ -5088,11 +5090,11 @@ find_first_nonascii(const unsigned char *start, const unsigned char *end)
             size_t u = load_unaligned(p, p2 - p) & ASCII_CHAR_MASK;
 #endif
             if (u) {
-                return p - start + (ctz(u) - 7) / 8;
+                return (ctz(u) - 7) / 8;
             }
             p = p2;
         }
-#else
+#else /* PY_LITTLE_ENDIAN && HAVE_CTZ */
         while (p < p2) {
             if (*p & 0x80) {
                 return p - start;