macOS/arm64 support, based on GH-21249

This is support for ctypes on macOS/arm64 based on PR 21249 by Lawrence D'Anna (Apple). Changes: - changed __builtin_available tests from 11.0 to 10.15 - added test to setup.py for ffi_closure_alloc and use that in malloc_closure.c - Minor change in the code path for ffi_prep_closure_var (coding style change)
ronaldoussoren · Nov 8, 2020 · Jul 20, 2020 · Jul 20, 2020 · Jul 20, 2020 · Jul 20, 2020
commit ea3c200890e1dc606745f6a5308725018595d001
diff --git a/Lib/test/test_bytes.py b/Lib/test/test_bytes.py
@@ -1035,6 +1035,7 @@ def test_from_format(self):
             c_char_p)
 
         PyBytes_FromFormat = pythonapi.PyBytes_FromFormat
+        PyBytes_FromFormat.argtypes = (c_char_p,)
         PyBytes_FromFormat.restype = py_object
 
         # basic tests

diff --git a/Lib/test/test_unicode.py b/Lib/test/test_unicode.py
@@ -2513,11 +2513,13 @@ class CAPITest(unittest.TestCase):
     def test_from_format(self):
         import_helper.import_module('ctypes')
         from ctypes import (
+            c_char_p,
             pythonapi, py_object, sizeof,
             c_int, c_long, c_longlong, c_ssize_t,
             c_uint, c_ulong, c_ulonglong, c_size_t, c_void_p)
         name = "PyUnicode_FromFormat"
         _PyUnicode_FromFormat = getattr(pythonapi, name)
+        _PyUnicode_FromFormat.argtypes = (c_char_p,)
         _PyUnicode_FromFormat.restype = py_object
 
         def PyUnicode_FromFormat(format, *args):

diff --git a/Modules/_ctypes/callbacks.c b/Modules/_ctypes/callbacks.c
@@ -1,6 +1,8 @@
 #include "Python.h"
 #include "frameobject.h"
 
+#include <stdbool.h>
+
 #include <ffi.h>
 #ifdef MS_WIN32
 #include <windows.h>
@@ -18,7 +20,7 @@ CThunkObject_dealloc(PyObject *myself)
     Py_XDECREF(self->callable);
     Py_XDECREF(self->restype);
     if (self->pcl_write)
-        ffi_closure_free(self->pcl_write);
+        Py_ffi_closure_free(self->pcl_write);
     PyObject_GC_Del(self);
 }
 
@@ -361,8 +363,7 @@ CThunkObject *_ctypes_alloc_callback(PyObject *callable,
 
     assert(CThunk_CheckExact((PyObject *)p));
 
-    p->pcl_write = ffi_closure_alloc(sizeof(ffi_closure),
-                                                                         &p->pcl_exec);
+    p->pcl_write = Py_ffi_closure_alloc(sizeof(ffi_closure), &p->pcl_exec);
     if (p->pcl_write == NULL) {
         PyErr_NoMemory();
         goto error;
@@ -408,13 +409,35 @@ CThunkObject *_ctypes_alloc_callback(PyObject *callable,
                      "ffi_prep_cif failed with %d", result);
         goto error;
     }
-#if defined(X86_DARWIN) || defined(POWERPC_DARWIN)
-    result = ffi_prep_closure(p->pcl_write, &p->cif, closure_fcn, p);
+#if HAVE_FFI_PREP_CLOSURE_LOC
+#   if USING_APPLE_OS_LIBFFI
+#      define HAVE_FFI_PREP_CLOSURE_LOC_RUNTIME __builtin_available(macos 10.15, ios 13, watchos 6, tvos 13, *)
+#   else
+#      define HAVE_FFI_PREP_CLOSURE_LOC_RUNTIME true
+#   endif
+    if (HAVE_FFI_PREP_CLOSURE_LOC_RUNTIME) {
+        result = ffi_prep_closure_loc(p->pcl_write, &p->cif, closure_fcn,
+                                    p,
+                                    p->pcl_exec);
+    } else
+#endif
+    {
+#if USING_APPLE_OS_LIBFFI && defined(__arm64__)
+        PyErr_Format(PyExc_NotImplementedError, "ffi_prep_closure_loc() is missing");
+        goto error;
 #else
-    result = ffi_prep_closure_loc(p->pcl_write, &p->cif, closure_fcn,
-                                  p,
-                                  p->pcl_exec);
+#ifdef MACOSX
+        #pragma clang diagnostic push
+        #pragma clang diagnostic ignored "-Wdeprecated-declarations"
 #endif
+        result = ffi_prep_closure(p->pcl_write, &p->cif, closure_fcn, p);
+
+#ifdef MACOSX
+        #pragma clang diagnostic pop
+#endif
+
+#endif
+    }
     if (result != FFI_OK) {
         PyErr_Format(PyExc_RuntimeError,
                      "ffi_prep_closure failed with %d", result);

diff --git a/Modules/_ctypes/callproc.c b/Modules/_ctypes/callproc.c
@@ -57,6 +57,8 @@
 #include "Python.h"
 #include "structmember.h"         // PyMemberDef
 
+#include <stdbool.h>
+
 #ifdef MS_WIN32
 #include <windows.h>
 #include <tchar.h>
@@ -812,7 +814,8 @@ static int _call_function_pointer(int flags,
                                   ffi_type **atypes,
                                   ffi_type *restype,
                                   void *resmem,
-                                  int argcount)
+                                  int argcount,
+                                  int argtypecount)
 {
     PyThreadState *_save = NULL; /* For Py_BLOCK_THREADS and Py_UNBLOCK_THREADS */
     PyObject *error_object = NULL;
@@ -835,14 +838,60 @@ static int _call_function_pointer(int flags,
     if ((flags & FUNCFLAG_CDECL) == 0)
         cc = FFI_STDCALL;
 #endif
-    if (FFI_OK != ffi_prep_cif(&cif,
-                               cc,
-                               argcount,
-                               restype,
-                               atypes)) {
-        PyErr_SetString(PyExc_RuntimeError,
-                        "ffi_prep_cif failed");
-        return -1;
+
+#   if USING_APPLE_OS_LIBFFI
+#      define HAVE_FFI_PREP_CIF_VAR_RUNTIME __builtin_available(macos 10.15, ios 13, watchos 6, tvos 13, *)
+#   elif HAVE_FFI_PREP_CIF_VAR
+#      define HAVE_FFI_PREP_CIF_VAR_RUNTIME true
+#   else
+#      define HAVE_FFI_PREP_CIF_VAR_RUNTIME false
+#   endif
+
+    /* Even on Apple-arm64 the calling convention for variadic functions conincides
+     * with the standard calling convention in the case that the function called
+     * only with its fixed arguments.   Thus, we do not need a special flag to be
+     * set on variadic functions.   We treat a function as variadic if it is called
+     * with a nonzero number of variadic arguments */
+    bool is_variadic = (argtypecount != 0 && argcount > argtypecount);
+    (void) is_variadic;
+
+#if defined(__APPLE__) && defined(__arm64__)
+    if (is_variadic) {
+        if (HAVE_FFI_PREP_CIF_VAR_RUNTIME) {
+        } else {
+            PyErr_SetString(PyExc_NotImplementedError, "ffi_prep_cif_var() is missing");
+            return -1;
+        }
+    }
+#endif
+
+#if HAVE_FFI_PREP_CIF_VAR
+    if (is_variadic) {
+        if (HAVE_FFI_PREP_CIF_VAR_RUNTIME) {
+            if (FFI_OK != ffi_prep_cif_var(&cif,
+                                        cc,
+                                        argtypecount,
+                                        argcount,
+                                        restype,
+                                        atypes)) {
+                PyErr_SetString(PyExc_RuntimeError,
+                                "ffi_prep_cif_var failed");
+                return -1;
+            }
+        }
+    } else
+#endif
+
+    {
+        if (FFI_OK != ffi_prep_cif(&cif,
+                                   cc,
+                                   argcount,
+                                   restype,
+                                   atypes)) {
+            PyErr_SetString(PyExc_RuntimeError,
+                            "ffi_prep_cif failed");
+            return -1;
+        }
     }
 
     if (flags & (FUNCFLAG_USE_ERRNO | FUNCFLAG_USE_LASTERROR)) {
@@ -1212,9 +1261,8 @@ PyObject *_ctypes_callproc(PPROC pProc,
 
     if (-1 == _call_function_pointer(flags, pProc, avalues, atypes,
                                      rtype, resbuf,
-                                     Py_SAFE_DOWNCAST(argcount,
-                                                      Py_ssize_t,
-                                                      int)))
+                                     Py_SAFE_DOWNCAST(argcount, Py_ssize_t, int),
+                                     Py_SAFE_DOWNCAST(argtype_count, Py_ssize_t, int)))
         goto cleanup;
 
 #ifdef WORDS_BIGENDIAN

diff --git a/Modules/_ctypes/ctypes.h b/Modules/_ctypes/ctypes.h
@@ -366,6 +366,14 @@ PyObject *_ctypes_get_errobj(int **pspace);
 extern PyObject *ComError;
 #endif
 
+#if USING_MALLOC_CLOSURE_DOT_C
+void Py_ffi_closure_free(void *p);
+void *Py_ffi_closure_alloc(size_t size, void** codeloc);
+#else
+#define Py_ffi_closure_free ffi_closure_free
+#define Py_ffi_closure_alloc ffi_closure_alloc
+#endif
+
 /*
  Local Variables:
  compile-command: "python setup.py -q build install --home ~"

diff --git a/Modules/_ctypes/malloc_closure.c b/Modules/_ctypes/malloc_closure.c
@@ -89,16 +89,27 @@ static void more_core(void)
 /******************************************************************/
 
 /* put the item back into the free list */
-void ffi_closure_free(void *p)
+void Py_ffi_closure_free(void *p)
 {
+#if USING_APPLE_OS_LIBFFI && HAVE_FFI_CLOSURE_ALLOC
+    if (__builtin_available(macos 10.15, ios 13, watchos 6, tvos 13, *)) {
+        ffi_closure_free(p);
+        return;
+    }
+#endif
     ITEM *item = (ITEM *)p;
     item->next = free_list;
     free_list = item;
 }
 
 /* return one item from the free list, allocating more if needed */
-void *ffi_closure_alloc(size_t ignored, void** codeloc)
+void *Py_ffi_closure_alloc(size_t size, void** codeloc)
 {
+#if USING_APPLE_OS_LIBFFI && HAVE_FFI_CLOSURE_ALLOC
+    if (__builtin_available(macos 10.15, ios 13, watchos 6, tvos 13, *)) {
+        return ffi_closure_alloc(size, codeloc);
+    }
+#endif
     ITEM *item;
     if (!free_list)
         more_core();

diff --git a/setup.py b/setup.py
@@ -229,6 +229,13 @@ def macosx_sdk_specified():
     macosx_sdk_root()
     return MACOS_SDK_SPECIFIED
 
+def is_macosx_at_least(vers):
+    if MACOS:
+        dep_target = sysconfig.get_config_var('MACOSX_DEPLOYMENT_TARGET')
+        if dep_target:
+            return tuple(map(int, dep_target.split('.'))) >= vers
+    return False
+
 
 def is_macosx_sdk_path(path):
     """
@@ -239,6 +246,13 @@ def is_macosx_sdk_path(path):
                 or path.startswith('/Library/') )
 
 
+def grep_headers_for(function, headers):
+    for header in headers:
+        with open(header, 'r') as f:
+            if function in f.read():
+                return True
+    return False
+
 def find_file(filename, std_dirs, paths):
     """Searches for the directory where a given file is located,
     and returns a possibly-empty list of additional directories, or None
@@ -2100,43 +2114,18 @@ def detect_tkinter(self):
                            library_dirs=added_lib_dirs))
         return True
 
-    def configure_ctypes_darwin(self, ext):
-        # Darwin (OS X) uses preconfigured files, in
-        # the Modules/_ctypes/libffi_osx directory.
-        ffi_srcdir = os.path.abspath(os.path.join(self.srcdir, 'Modules',
-                                                  '_ctypes', 'libffi_osx'))
-        sources = [os.path.join(ffi_srcdir, p)
-                   for p in ['ffi.c',
-                             'x86/darwin64.S',
-                             'x86/x86-darwin.S',
-                             'x86/x86-ffi_darwin.c',
-                             'x86/x86-ffi64.c',
-                             'powerpc/ppc-darwin.S',
-                             'powerpc/ppc-darwin_closure.S',
-                             'powerpc/ppc-ffi_darwin.c',
-                             'powerpc/ppc64-darwin_closure.S',
-                             ]]
-
-        # Add .S (preprocessed assembly) to C compiler source extensions.
-        self.compiler.src_extensions.append('.S')
-
-        include_dirs = [os.path.join(ffi_srcdir, 'include'),
-                        os.path.join(ffi_srcdir, 'powerpc')]
-        ext.include_dirs.extend(include_dirs)
-        ext.sources.extend(sources)
-        return True
-
     def configure_ctypes(self, ext):
-        if not self.use_system_libffi:
-            if MACOS:
-                return self.configure_ctypes_darwin(ext)
-            print('INFO: Could not locate ffi libs and/or headers')
-            return False
         return True
 
     def detect_ctypes(self):
         # Thomas Heller's _ctypes module
-        self.use_system_libffi = False
+
+        if (not sysconfig.get_config_var("LIBFFI_INCLUDEDIR") and MACOS and
+            (is_macosx_at_least((10,15)) or '-arch arm64' in sysconfig.get_config_var("CFLAGS"))):
+            self.use_system_libffi = True
+        else:
+            self.use_system_libffi = '--with-system-ffi' in sysconfig.get_config_var("CONFIG_ARGS")
+
         include_dirs = []
         extra_compile_args = ['-DPy_BUILD_CORE_MODULE']
         extra_link_args = []
@@ -2149,11 +2138,10 @@ def detect_ctypes(self):
 
         if MACOS:
             sources.append('_ctypes/malloc_closure.c')
-            sources.append('_ctypes/darwin/dlfcn_simple.c')
+            extra_compile_args.append('-DUSING_MALLOC_CLOSURE_DOT_C=1')
+            #sources.append('_ctypes/darwin/dlfcn_simple.c')
             extra_compile_args.append('-DMACOSX')
             include_dirs.append('_ctypes/darwin')
-            # XXX Is this still needed?
-            # extra_link_args.extend(['-read_only_relocs', 'warning'])
 
         elif HOST_PLATFORM == 'sunos5':
             # XXX This shouldn't be necessary; it appears that some
@@ -2183,31 +2171,47 @@ def detect_ctypes(self):
                                sources=['_ctypes/_ctypes_test.c'],
                                libraries=['m']))
 
+        ffi_inc = sysconfig.get_config_var("LIBFFI_INCLUDEDIR")
+        ffi_lib = None
+
         ffi_inc_dirs = self.inc_dirs.copy()
         if MACOS:
-            if '--with-system-ffi' not in sysconfig.get_config_var("CONFIG_ARGS"):
-                return
-            # OS X 10.5 comes with libffi.dylib; the include files are
-            # in /usr/include/ffi
-            ffi_inc_dirs.append('/usr/include/ffi')
-
-        ffi_inc = [sysconfig.get_config_var("LIBFFI_INCLUDEDIR")]
-        if not ffi_inc or ffi_inc[0] == '':
-            ffi_inc = find_file('ffi.h', [], ffi_inc_dirs)
-        if ffi_inc is not None:
-            ffi_h = ffi_inc[0] + '/ffi.h'
+            # XXX: The define should only be added when actually using the system
+            #      version (and not a locally compiled one)
+            ext.extra_compile_args.append("-DUSING_APPLE_OS_LIBFFI=1")
+            ffi_in_sdk = os.path.join(macosx_sdk_root(), "usr/include/ffi")
+            if os.path.exists(ffi_in_sdk):
+                ffi_inc = ffi_in_sdk
+                ffi_lib = 'ffi'
+            else:
+                # OS X 10.5 comes with libffi.dylib; the include files are
+                # in /usr/include/ffi
+                ffi_inc_dirs.append('/usr/include/ffi')
+
+        if not ffi_inc:
+            found = find_file('ffi.h', [], ffi_inc_dirs)
+            if found:
+                ffi_inc = found[0]
+        if ffi_inc:
+            ffi_h = ffi_inc + '/ffi.h'
             if not os.path.exists(ffi_h):
                 ffi_inc = None
                 print('Header file {} does not exist'.format(ffi_h))
-        ffi_lib = None
-        if ffi_inc is not None:
+        if ffi_lib is None and ffi_inc:
             for lib_name in ('ffi', 'ffi_pic'):
                 if (self.compiler.find_library_file(self.lib_dirs, lib_name)):
                     ffi_lib = lib_name
                     break
 
         if ffi_inc and ffi_lib:
-            ext.include_dirs.extend(ffi_inc)
+            ffi_headers = glob(os.path.join(ffi_inc, '*.h'))
+            if grep_headers_for('ffi_prep_cif_var', ffi_headers):
+                ext.extra_compile_args.append("-DHAVE_FFI_PREP_CIF_VAR=1")
+            if grep_headers_for('ffi_prep_closure_loc', ffi_headers):
+                ext.extra_compile_args.append("-DHAVE_FFI_PREP_CLOSURE_LOC=1")
+            if grep_headers_for('ffi_closure_alloc', ffi_headers):
+                ext.extra_compile_args.append("-DHAVE_FFI_CLOSURE_ALLOC=1")
+            ext.include_dirs.append(ffi_inc)
             ext.libraries.append(ffi_lib)
             self.use_system_libffi = True