Apply Chris' changes

Co-Authored-By: Chris Eibl <138194463+chris-eibl@users.noreply.github.com>

Author: Fidget-Spinner

Include/internal/pycore_ceval.h

@@ -391,28 +391,6 @@ _PyForIter_VirtualIteratorNext(PyThreadState* tstate, struct _PyInterpreterFrame
 #define SPECIAL___AEXIT__   3
 #define SPECIAL_MAX   3
 
-// Special counterparts of ceval functions for performance reasons
-PyAPI_FUNC(int) _PyEval_Mapping_GetOptionalItem(PyObject *obj, PyObject *key, PyObject **result);
-
-#if defined(_MSC_VER) && !defined(__clang__) && _Py_TAIL_CALL_INTERP
-#  define Py_NO_INLINE_MSVC_TAILCALL Py_NO_INLINE
-#else
-#  define Py_NO_INLINE_MSVC_TAILCALL
-#endif
-
-// Tells the compiler that this variable cannot be alised.
-#if defined(_MSC_VER) && !defined(__clang__)
-#  define Py_UNALIASED(var) restrict var
-#else
-#  define Py_UNALIASED(var) var
-#endif
-
-// Just a scope. Hints to the programmer and compiler
-// That any local variable defined within this block MUST
-// not escape from the current definition.
-# define Py_BEGIN_LOCALS_MUST_NOT_ESCAPE {
-# define Py_END_LOCALS_MUST_NOT_ESCAPE }
-
 #ifdef __cplusplus
 }
 #endif

Objects/abstract.c

@@ -205,8 +205,14 @@ PyObject_GetItem(PyObject *o, PyObject *key)
     return type_error("'%.200s' object is not subscriptable", o);
 }
 
+// MSVC fails during a tail call release build with loads of
+// error C4737: Unable to perform required tail call.
+// without using Py_NO_INLINE here, but PGO works fine.
+#if defined(_MSC_VER) && !defined(__clang__) && _Py_TAIL_CALL_INTERP && !defined(_Py_USING_PGO)
+Py_NO_INLINE
+#endif
 int
-PyMapping_GetOptionalItem(PyObject *obj, PyObject *key, PyObject **result)
+PyMapping_GetOptionalItem(PyObject *obj, PyObject *key, PyObject **restrict result)
 {
     if (PyDict_CheckExact(obj)) {
         return PyDict_GetItemRef(obj, key, result);
@@ -224,12 +230,6 @@ PyMapping_GetOptionalItem(PyObject *obj, PyObject *key, PyObject **result)
     return 0;
 }
 
-Py_NO_INLINE_MSVC_TAILCALL int
-_PyEval_Mapping_GetOptionalItem(PyObject *obj, PyObject *key, PyObject **Py_UNALIASED(result))
-{
-    return PyMapping_GetOptionalItem(obj, key, result);
-}
-
 int
 PyObject_SetItem(PyObject *o, PyObject *key, PyObject *value)
 {

Objects/dictobject.c

@@ -2225,7 +2225,7 @@ _PyDict_NewPresized(Py_ssize_t minused)
     return dict_new_presized(minused, false);
 }
 
-Py_NO_INLINE_MSVC_TAILCALL PyObject *
+PyObject *
 _PyDict_FromItems(PyObject *const *keys, Py_ssize_t keys_offset,
                   PyObject *const *values, Py_ssize_t values_offset,
                   Py_ssize_t length)

Python/bytecodes.c

@@ -1535,7 +1535,7 @@ dummy_func(
 
         inst(LOAD_BUILD_CLASS, ( -- bc)) {
             PyObject *bc_o;
-            int err = _PyEval_Mapping_GetOptionalItem(BUILTINS(), &_Py_ID(__build_class__), &bc_o);
+            int err = PyMapping_GetOptionalItem(BUILTINS(), &_Py_ID(__build_class__), &bc_o);
             ERROR_IF(err < 0);
             if (bc_o == NULL) {
                 _PyErr_SetString(tstate, PyExc_NameError,
@@ -1739,7 +1739,7 @@ dummy_func(
         inst(LOAD_FROM_DICT_OR_GLOBALS, (mod_or_class_dict -- v)) {
             PyObject *name = GETITEM(FRAME_CO_NAMES, oparg);
             PyObject *v_o;
-            int err = _PyEval_Mapping_GetOptionalItem(PyStackRef_AsPyObjectBorrow(mod_or_class_dict), name, &v_o);
+            int err = PyMapping_GetOptionalItem(PyStackRef_AsPyObjectBorrow(mod_or_class_dict), name, &v_o);
             PyStackRef_CLOSE(mod_or_class_dict);
             ERROR_IF(err < 0);
             if (v_o == NULL) {
@@ -1762,11 +1762,11 @@ dummy_func(
                 else {
                     /* Slow-path if globals or builtins is not a dict */
                     /* namespace 1: globals */
-                    int err = _PyEval_Mapping_GetOptionalItem(GLOBALS(), name, &v_o);
+                    int err = PyMapping_GetOptionalItem(GLOBALS(), name, &v_o);
                     ERROR_IF(err < 0);
                     if (v_o == NULL) {
                         /* namespace 2: builtins */
-                        int err = _PyEval_Mapping_GetOptionalItem(BUILTINS(), name, &v_o);
+                        int err = PyMapping_GetOptionalItem(BUILTINS(), name, &v_o);
                         ERROR_IF(err < 0);
                         if (v_o == NULL) {
                             _PyEval_FormatExcCheckArg(
@@ -1932,7 +1932,7 @@ dummy_func(
             assert(class_dict);
             assert(oparg >= 0 && oparg < _PyFrame_GetCode(frame)->co_nlocalsplus);
             name = PyTuple_GET_ITEM(_PyFrame_GetCode(frame)->co_localsplusnames, oparg);
-            int err = _PyEval_Mapping_GetOptionalItem(class_dict, name, &value_o);
+            int err = PyMapping_GetOptionalItem(class_dict, name, &value_o);
             if (err < 0) {
                 ERROR_NO_POP();
             }
@@ -2122,7 +2122,7 @@ dummy_func(
                 ERROR_IF(true);
             }
             /* check if __annotations__ in locals()... */
-            int err = _PyEval_Mapping_GetOptionalItem(LOCALS(), &_Py_ID(__annotations__), &ann_dict);
+            int err = PyMapping_GetOptionalItem(LOCALS(), &_Py_ID(__annotations__), &ann_dict);
             ERROR_IF(err < 0);
             if (ann_dict == NULL) {
                 ann_dict = PyDict_New();
@@ -2227,10 +2227,10 @@ dummy_func(
             // we make no attempt to optimize here; specializations should
             // handle any case whose performance we care about
             PyObject *super;
-            Py_BEGIN_LOCALS_MUST_NOT_ESCAPE;
-            PyObject *stack[] = {class, self};
-            super = PyObject_Vectorcall(global_super, stack, oparg & 2, NULL);
-            Py_END_LOCALS_MUST_NOT_ESCAPE;
+            {
+                PyObject *stack[] = {class, self};
+                super = PyObject_Vectorcall(global_super, stack, oparg & 2, NULL);
+            }
             if (opcode == INSTRUMENTED_LOAD_SUPER_ATTR) {
                 PyObject *arg = oparg & 2 ? class : &_PyInstrumentation_MISSING;
                 if (super == NULL) {
@@ -2290,11 +2290,11 @@ dummy_func(
             PyTypeObject *cls = (PyTypeObject *)class;
             int method_found = 0;
             PyObject *attr_o;
-            Py_BEGIN_LOCALS_MUST_NOT_ESCAPE;
-            int *method_found_ptr = &method_found;
-            attr_o = _PySuper_Lookup(cls, self, name,
-                                Py_TYPE(self)->tp_getattro == PyObject_GenericGetAttr ? method_found_ptr : NULL);
-            Py_END_LOCALS_MUST_NOT_ESCAPE;
+            {
+                int *method_found_ptr = &method_found;
+                attr_o = _PySuper_Lookup(cls, self, name,
+                    Py_TYPE(self)->tp_getattro == PyObject_GenericGetAttr ? method_found_ptr : NULL);
+            }
             if (attr_o == NULL) {
                 ERROR_NO_POP();
             }
@@ -3518,12 +3518,12 @@ dummy_func(
             assert(PyStackRef_IsTaggedInt(lasti));
             (void)lasti; // Shut up compiler warning if asserts are off
             PyObject* res_o;
-            Py_BEGIN_LOCALS_MUST_NOT_ESCAPE;
-            PyObject *stack[5] = {NULL, PyStackRef_AsPyObjectBorrow(exit_self), exc, val_o, tb};
-            int has_self = !PyStackRef_IsNull(exit_self);
-            res_o = PyObject_Vectorcall(exit_func_o, stack + 2 - has_self,
-                    (3 + has_self) | PY_VECTORCALL_ARGUMENTS_OFFSET, NULL);
-            Py_END_LOCALS_MUST_NOT_ESCAPE;
+            {
+                PyObject *stack[5] = {NULL, PyStackRef_AsPyObjectBorrow(exit_self), exc, val_o, tb};
+                int has_self = !PyStackRef_IsNull(exit_self);
+                res_o = PyObject_Vectorcall(exit_func_o, stack + 2 - has_self,
+                        (3 + has_self) | PY_VECTORCALL_ARGUMENTS_OFFSET, NULL);
+            }
             Py_XDECREF(original_tb);
             ERROR_IF(res_o == NULL);
             res = PyStackRef_FromPyObjectSteal(res_o);