bpo-40302: Add _Py_bswap32() function to pyport.h

Include/internal/pycore_byteswap.h

@@ -0,0 +1,89 @@
+/* Bytes swap functions, reverse order of bytes:
+
+   - _Py_bswap16(uint16_t)
+   - _Py_bswap32(uint32_t)
+   - _Py_bswap64(uint64_t)
+*/
+
+#ifndef Py_INTERNAL_BSWAP_H
+#define Py_INTERNAL_BSWAP_H
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifndef Py_BUILD_CORE
+#  error "this header requires Py_BUILD_CORE define"
+#endif
+
+#if defined(__clang__) || \
+    (defined(__GNUC__) && \
+     ((__GNUC__ >= 5) || (__GNUC__ == 4) && (__GNUC_MINOR__ >= 8)))
+   /* __builtin_bswap16() is available since GCC 4.8,
+      __builtin_bswap32() is available since GCC 4.3,
+      __builtin_bswap64() is available since GCC 4.3. */
+#  define _PY_HAVE_BUILTIN_BSWAP
+#endif
+
+#ifdef _MSC_VER
+   /* Get _byteswap_ushort(), _byteswap_ulong(), _byteswap_uint64() */
+#  include <intrin.h>
+#endif
+
+static inline uint16_t
+_Py_bswap16(uint16_t word)
+{
+#ifdef _PY_HAVE_BUILTIN_BSWAP
+    return __builtin_bswap16(word);
+#elif defined(_MSC_VER)
+    Py_BUILD_ASSERT(sizeof(word) == sizeof(unsigned short));
+    return _byteswap_ushort(word);
+#else
+    // Portable implementation which doesn't rely on circular bit shift
+    return ( ((word & UINT16_C(0x00FF)) << 8)
+           | ((word & UINT16_C(0xFF00)) >> 8));
+#endif
+}
+
+static inline uint32_t
+_Py_bswap32(uint32_t word)
+{
+#ifdef _PY_HAVE_BUILTIN_BSWAP
+    return __builtin_bswap32(word);
+#elif defined(_MSC_VER)
+    Py_BUILD_ASSERT(sizeof(word) == sizeof(unsigned long));
+    return _byteswap_ulong(word);
+#else
+    // Portable implementation which doesn't rely on circular bit shift
+    return ( ((word & UINT32_C(0x000000FF)) << 24)
+           | ((word & UINT32_C(0x0000FF00)) <<  8)
+           | ((word & UINT32_C(0x00FF0000)) >>  8)
+           | ((word & UINT32_C(0xFF000000)) >> 24));
+#endif
+}
+
+static inline uint64_t
+_Py_bswap64(uint64_t word)
+{
+#ifdef _PY_HAVE_BUILTIN_BSWAP
+    return __builtin_bswap64(word);
+#elif defined(_MSC_VER)
+    return _byteswap_uint64(word);
+#else
+    // Portable implementation which doesn't rely on circular bit shift
+    return ( ((word & UINT64_C(0x00000000000000FF)) << 56)
+           | ((word & UINT64_C(0x000000000000FF00)) << 40)
+           | ((word & UINT64_C(0x0000000000FF0000)) << 24)
+           | ((word & UINT64_C(0x00000000FF000000)) <<  8)
+           | ((word & UINT64_C(0x000000FF00000000)) >>  8)
+           | ((word & UINT64_C(0x0000FF0000000000)) >> 24)
+           | ((word & UINT64_C(0x00FF000000000000)) >> 40)
+           | ((word & UINT64_C(0xFF00000000000000)) >> 56));
+#endif
+}
+
+
+#ifdef __cplusplus
+}
+#endif
+#endif /* !Py_INTERNAL_BSWAP_H */
+

Include/pyport.h

@@ -768,11 +768,11 @@ extern char * _getpty(int *, int, mode_t, int);
  */
 
 #ifdef WORDS_BIGENDIAN
-#define PY_BIG_ENDIAN 1
-#define PY_LITTLE_ENDIAN 0
+#  define PY_BIG_ENDIAN 1
+#  define PY_LITTLE_ENDIAN 0
 #else
-#define PY_BIG_ENDIAN 0
-#define PY_LITTLE_ENDIAN 1
+#  define PY_BIG_ENDIAN 0
+#  define PY_LITTLE_ENDIAN 1
 #endif
 
 #ifdef Py_BUILD_CORE

Lib/test/test_capi.py

@@ -24,6 +24,8 @@
 # Skip this test if the _testcapi module isn't available.
 _testcapi = support.import_module('_testcapi')
 
+import _testinternalcapi
+
 # Were we compiled --with-pydebug or with #define Py_DEBUG?
 Py_DEBUG = hasattr(sys, 'gettotalrefcount')
 
@@ -658,6 +660,12 @@ class Test_testcapi(unittest.TestCase):
                     if name.startswith('test_') and not name.endswith('_code'))
 
 
+class Test_testinternalcapi(unittest.TestCase):
+    locals().update((name, getattr(_testinternalcapi, name))
+                    for name in dir(_testinternalcapi)
+                    if name.startswith('test_'))
+
+
 class PyMemDebugTests(unittest.TestCase):
     PYTHONMALLOC = 'debug'
     # '0x04c06e0' or '04C06E0'

Makefile.pre.in

@@ -1083,6 +1083,7 @@ PYTHON_HEADERS= \
 		$(srcdir)/Include/internal/pycore_abstract.h \
 		$(srcdir)/Include/internal/pycore_accu.h \
 		$(srcdir)/Include/internal/pycore_atomic.h \
+		$(srcdir)/Include/internal/pycore_byteswap.h \
 		$(srcdir)/Include/internal/pycore_bytes_methods.h \
 		$(srcdir)/Include/internal/pycore_call.h \
 		$(srcdir)/Include/internal/pycore_ceval.h \

Modules/Setup

@@ -247,8 +247,8 @@ _symtable symtablemodule.c
 # The _sha module implements the SHA checksum algorithms.
 # (NIST's Secure Hash Algorithms.)
 #_sha1 sha1module.c
-#_sha256 sha256module.c
-#_sha512 sha512module.c
+#_sha256 sha256module.c -DPy_BUILD_CORE_BUILTIN
+#_sha512 sha512module.c -DPy_BUILD_CORE_BUILTIN
 #_sha3 _sha3/sha3module.c
 
 # _blake module

Modules/_ctypes/cfield.c

@@ -1,4 +1,5 @@
 #include "Python.h"
+#include "pycore_byteswap.h"      // _Py_bswap32()
 
 #include <ffi.h>
 #ifdef MS_WIN32
@@ -448,46 +449,32 @@ get_ulonglong(PyObject *v, unsigned long long *p)
      ( ( (type)x & ~(BIT_MASK(type, size) << LOW_BIT(size)) ) | ( ((type)v & BIT_MASK(type, size)) << LOW_BIT(size) ) ) \
      : (type)v)
 
-/* byte swapping macros */
-#define SWAP_2(v)                               \
-    ( ( (v >> 8) & 0x00FF) |                    \
-      ( (v << 8) & 0xFF00) )
-
-#define SWAP_4(v)                       \
-    ( ( (v & 0x000000FF) << 24 ) |  \
-      ( (v & 0x0000FF00) <<  8 ) |  \
-      ( (v & 0x00FF0000) >>  8 ) |  \
-      ( ((v >> 24) & 0xFF)) )
-
-#ifdef _MSC_VER
-#define SWAP_8(v)                               \
-    ( ( (v & 0x00000000000000FFL) << 56 ) |  \
-      ( (v & 0x000000000000FF00L) << 40 ) |  \
-      ( (v & 0x0000000000FF0000L) << 24 ) |  \
-      ( (v & 0x00000000FF000000L) <<  8 ) |  \
-      ( (v & 0x000000FF00000000L) >>  8 ) |  \
-      ( (v & 0x0000FF0000000000L) >> 24 ) |  \
-      ( (v & 0x00FF000000000000L) >> 40 ) |  \
-      ( ((v >> 56) & 0xFF)) )
+#if SIZEOF_SHORT == 2
+#  define SWAP_SHORT _Py_bswap16
 #else
-#define SWAP_8(v)                               \
-    ( ( (v & 0x00000000000000FFLL) << 56 ) |  \
-      ( (v & 0x000000000000FF00LL) << 40 ) |  \
-      ( (v & 0x0000000000FF0000LL) << 24 ) |  \
-      ( (v & 0x00000000FF000000LL) <<  8 ) |  \
-      ( (v & 0x000000FF00000000LL) >>  8 ) |  \
-      ( (v & 0x0000FF0000000000LL) >> 24 ) |  \
-      ( (v & 0x00FF000000000000LL) >> 40 ) |  \
-      ( ((v >> 56) & 0xFF)) )
+#  error "unsupported short size"
 #endif
 
-#define SWAP_INT SWAP_4
+#if SIZEOF_INT == 4
+#  define SWAP_INT _Py_bswap32
+#else
+#  error "unsupported int size"
+#endif
 
 #if SIZEOF_LONG == 4
-# define SWAP_LONG SWAP_4
+#  define SWAP_LONG _Py_bswap32
 #elif SIZEOF_LONG == 8
-# define SWAP_LONG SWAP_8
+#  define SWAP_LONG _Py_bswap64
+#else
+#  error "unsupported long size"
+#endif
+
+#if SIZEOF_LONG_LONG == 8
+#  define SWAP_LONG_LONG _Py_bswap64
+#else
+#  error "unsupported long long size"
 #endif
+
 /*****************************************************************
  * The setter methods return an object which must be kept alive, to keep the
  * data valid which has been stored in the memory block.  The ctypes object
@@ -569,12 +556,13 @@ h_set_sw(void *ptr, PyObject *value, Py_ssize_t size)
 {
     long val;
     short field;
-    if (get_long(value, &val) < 0)
+    if (get_long(value, &val) < 0) {
         return NULL;
+    }
     memcpy(&field, ptr, sizeof(field));
-    field = SWAP_2(field);
+    field = SWAP_SHORT(field);
     field = SET(short, field, val, size);
-    field = SWAP_2(field);
+    field = SWAP_SHORT(field);
     memcpy(ptr, &field, sizeof(field));
     _RET(value);
 }
@@ -593,7 +581,7 @@ h_get_sw(void *ptr, Py_ssize_t size)
 {
     short val;
     memcpy(&val, ptr, sizeof(val));
-    val = SWAP_2(val);
+    val = SWAP_SHORT(val);
     GET_BITFIELD(val, size);
     return PyLong_FromLong(val);
 }
@@ -616,12 +604,13 @@ H_set_sw(void *ptr, PyObject *value, Py_ssize_t size)
 {
     unsigned long val;
     unsigned short field;
-    if (get_ulong(value, &val) < 0)
+    if (get_ulong(value, &val) < 0) {
         return NULL;
+    }
     memcpy(&field, ptr, sizeof(field));
-    field = SWAP_2(field);
+    field = SWAP_SHORT(field);
     field = SET(unsigned short, field, val, size);
-    field = SWAP_2(field);
+    field = SWAP_SHORT(field);
     memcpy(ptr, &field, sizeof(field));
     _RET(value);
 }
@@ -641,7 +630,7 @@ H_get_sw(void *ptr, Py_ssize_t size)
 {
     unsigned short val;
     memcpy(&val, ptr, sizeof(val));
-    val = SWAP_2(val);
+    val = SWAP_SHORT(val);
     GET_BITFIELD(val, size);
     return PyLong_FromLong(val);
 }
@@ -664,8 +653,9 @@ i_set_sw(void *ptr, PyObject *value, Py_ssize_t size)
 {
     long val;
     int field;
-    if (get_long(value, &val) < 0)
+    if (get_long(value, &val) < 0) {
         return NULL;
+    }
     memcpy(&field, ptr, sizeof(field));
     field = SWAP_INT(field);
     field = SET(int, field, val, size);
@@ -757,8 +747,9 @@ I_set_sw(void *ptr, PyObject *value, Py_ssize_t size)
 {
     unsigned long val;
     unsigned int field;
-    if (get_ulong(value, &val) < 0)
+    if (get_ulong(value, &val) < 0) {
         return  NULL;
+    }
     memcpy(&field, ptr, sizeof(field));
     field = SWAP_INT(field);
     field = SET(unsigned int, field, (unsigned int)val, size);
@@ -805,8 +796,9 @@ l_set_sw(void *ptr, PyObject *value, Py_ssize_t size)
 {
     long val;
     long field;
-    if (get_long(value, &val) < 0)
+    if (get_long(value, &val) < 0) {
         return NULL;
+    }
     memcpy(&field, ptr, sizeof(field));
     field = SWAP_LONG(field);
     field = SET(long, field, val, size);
@@ -853,8 +845,9 @@ L_set_sw(void *ptr, PyObject *value, Py_ssize_t size)
 {
     unsigned long val;
     unsigned long field;
-    if (get_ulong(value, &val) < 0)
+    if (get_ulong(value, &val) < 0) {
         return  NULL;
+    }
     memcpy(&field, ptr, sizeof(field));
     field = SWAP_LONG(field);
     field = SET(unsigned long, field, val, size);
@@ -901,12 +894,13 @@ q_set_sw(void *ptr, PyObject *value, Py_ssize_t size)
 {
     long long val;
     long long field;
-    if (get_longlong(value, &val) < 0)
+    if (get_longlong(value, &val) < 0) {
         return NULL;
+    }
     memcpy(&field, ptr, sizeof(field));
-    field = SWAP_8(field);
+    field = SWAP_LONG_LONG(field);
     field = SET(long long, field, val, size);
-    field = SWAP_8(field);
+    field = SWAP_LONG_LONG(field);
     memcpy(ptr, &field, sizeof(field));
     _RET(value);
 }
@@ -925,7 +919,7 @@ q_get_sw(void *ptr, Py_ssize_t size)
 {
     long long val;
     memcpy(&val, ptr, sizeof(val));
-    val = SWAP_8(val);
+    val = SWAP_LONG_LONG(val);
     GET_BITFIELD(val, size);
     return PyLong_FromLongLong(val);
 }
@@ -948,12 +942,13 @@ Q_set_sw(void *ptr, PyObject *value, Py_ssize_t size)
 {
     unsigned long long val;
     unsigned long long field;
-    if (get_ulonglong(value, &val) < 0)
+    if (get_ulonglong(value, &val) < 0) {
         return NULL;
+    }
     memcpy(&field, ptr, sizeof(field));
-    field = SWAP_8(field);
+    field = SWAP_LONG_LONG(field);
     field = SET(unsigned long long, field, val, size);
-    field = SWAP_8(field);
+    field = SWAP_LONG_LONG(field);
     memcpy(ptr, &field, sizeof(field));
     _RET(value);
 }
@@ -972,7 +967,7 @@ Q_get_sw(void *ptr, Py_ssize_t size)
 {
     unsigned long long val;
     memcpy(&val, ptr, sizeof(val));
-    val = SWAP_8(val);
+    val = SWAP_LONG_LONG(val);
     GET_BITFIELD(val, size);
     return PyLong_FromUnsignedLongLong(val);
 }