bpo-43683: Streamline YIELD_VALUE and SEND (GH-30723)

* Split YIELD_VALUE into ASYNC_GEN_WRAP; YIELD_VALUE for async generators.

* Split SEND into SEND; YIELD_VALUE.

* Document new opcodes.

9 files changed, 72 insertions, 43 deletions

diff --git a/Doc/library/dis.rst b/Doc/library/dis.rst
index c9a4768..ddba668 100644
--- a/Doc/library/dis.rst
+++ b/Doc/library/dis.rst
@@ -1233,6 +1233,22 @@ All of the following opcodes use their arguments.
     .. versionadded:: 3.11
 
 
+.. opcode:: SEND
+
+    Sends ``None`` to the sub-generator of this generator.
+    Used in ``yield from`` and ``await`` statements.
+
+    .. versionadded:: 3.11
+
+
+.. opcode:: ASYNC_GEN_WRAP
+
+    Wraps the value on top of the stack in an ``async_generator_wrapped_value``.
+    Used to yield in async generators.
+
+    .. versionadded:: 3.11
+
+
 .. opcode:: HAVE_ARGUMENT
 
    This is not really an opcode.  It identifies the dividing line between
diff --git a/Include/opcode.h b/Include/opcode.h
index c0686bd..985758d 100644
--- a/Include/opcode.h
+++ b/Include/opcode.h
@@ -44,6 +44,7 @@ extern "C" {
 #define IMPORT_STAR                      84
 #define SETUP_ANNOTATIONS                85
 #define YIELD_VALUE                      86
+#define ASYNC_GEN_WRAP                   87
 #define PREP_RERAISE_STAR                88
 #define POP_EXCEPT                       89
 #define HAVE_ARGUMENT                    90
@@ -165,12 +166,12 @@ extern "C" {
 #define STORE_ATTR_ADAPTIVE              79
 #define STORE_ATTR_INSTANCE_VALUE        80
 #define STORE_ATTR_SLOT                  81
-#define STORE_ATTR_WITH_HINT             87
-#define LOAD_FAST__LOAD_FAST            131
-#define STORE_FAST__LOAD_FAST           140
-#define LOAD_FAST__LOAD_CONST           141
-#define LOAD_CONST__LOAD_FAST           143
-#define STORE_FAST__STORE_FAST          150
+#define STORE_ATTR_WITH_HINT            131
+#define LOAD_FAST__LOAD_FAST            140
+#define STORE_FAST__LOAD_FAST           141
+#define LOAD_FAST__LOAD_CONST           143
+#define LOAD_CONST__LOAD_FAST           150
+#define STORE_FAST__STORE_FAST          153
 #define DO_TRACING                      255
 #ifdef NEED_OPCODE_JUMP_TABLES
 static uint32_t _PyOpcode_RelativeJump[8] = {
diff --git a/Lib/importlib/_bootstrap_external.py b/Lib/importlib/_bootstrap_external.py
index 1560e60..cd4f69c 100644
--- a/Lib/importlib/_bootstrap_external.py
+++ b/Lib/importlib/_bootstrap_external.py
@@ -381,6 +381,7 @@ _code_type = type(_write_atomic.__code__)
 #     Python 3.11a4 3473 (Add POP_JUMP_IF_NOT_NONE/POP_JUMP_IF_NONE opcodes)
 #     Python 3.11a4 3474 (Add RESUME opcode)
 #     Python 3.11a5 3475 (Add RETURN_GENERATOR opcode)
+#     Python 3.11a5 3476 (Add ASYNC_GEN_WRAP opcode)
 
 #     Python 3.12 will start with magic number 3500
 
@@ -394,7 +395,7 @@ _code_type = type(_write_atomic.__code__)
 # Whenever MAGIC_NUMBER is changed, the ranges in the magic_values array
 # in PC/launcher.c must also be updated.
 
-MAGIC_NUMBER = (3475).to_bytes(2, 'little') + b'\r\n'
+MAGIC_NUMBER = (3476).to_bytes(2, 'little') + b'\r\n'
 _RAW_MAGIC_NUMBER = int.from_bytes(MAGIC_NUMBER, 'little')  # For import.c
 
 _PYCACHE = '__pycache__'
diff --git a/Lib/opcode.py b/Lib/opcode.py
index 73b41d2..1bd48ee 100644
--- a/Lib/opcode.py
+++ b/Lib/opcode.py
@@ -101,7 +101,7 @@ def_op('RETURN_VALUE', 83)
 def_op('IMPORT_STAR', 84)
 def_op('SETUP_ANNOTATIONS', 85)
 def_op('YIELD_VALUE', 86)
-
+def_op('ASYNC_GEN_WRAP', 87)
 def_op('PREP_RERAISE_STAR', 88)
 def_op('POP_EXCEPT', 89)
 
diff --git a/Misc/NEWS.d/next/Core and Builtins/2022-01-20-17-13-49.bpo-43683.BqQ26Z.rst b/Misc/NEWS.d/next/Core and Builtins/2022-01-20-17-13-49.bpo-43683.BqQ26Z.rst
new file mode 100644
index 0000000..737f44f
--- /dev/null
+++ b/Misc/NEWS.d/next/Core and Builtins/2022-01-20-17-13-49.bpo-43683.BqQ26Z.rst
@@ -0,0 +1,3 @@
+Add ASYNC_GEN_WRAP opcode to wrap the value to be yielded in async
+generators. Removes the need to special case async generators in the
+``YIELD_VALUE`` instruction.
diff --git a/Objects/genobject.c b/Objects/genobject.c
index 46b0190..b2d402e 100644
--- a/Objects/genobject.c
+++ b/Objects/genobject.c
@@ -353,7 +353,7 @@ _PyGen_yf(PyGenObject *gen)
         PyObject *bytecode = gen->gi_code->co_code;
         unsigned char *code = (unsigned char *)PyBytes_AS_STRING(bytecode);
 
-        if (frame->f_lasti < 0) {
+        if (frame->f_lasti < 1) {
             /* Return immediately if the frame didn't start yet. SEND
                always come after LOAD_CONST: a code object should not start
                with SEND */
@@ -361,7 +361,7 @@ _PyGen_yf(PyGenObject *gen)
             return NULL;
         }
 
-        if (code[frame->f_lasti*sizeof(_Py_CODEUNIT)] != SEND || frame->stacktop < 0)
+        if (code[(frame->f_lasti-1)*sizeof(_Py_CODEUNIT)] != SEND || frame->stacktop < 0)
             return NULL;
         yf = _PyFrame_StackPeek(frame);
         Py_INCREF(yf);
@@ -488,6 +488,8 @@ _gen_throw(PyGenObject *gen, int close_on_genexit,
             assert(frame->f_lasti >= 0);
             PyObject *bytecode = gen->gi_code->co_code;
             unsigned char *code = (unsigned char *)PyBytes_AS_STRING(bytecode);
+            /* Backup to SEND */
+            frame->f_lasti--;
             assert(code[frame->f_lasti*sizeof(_Py_CODEUNIT)] == SEND);
             int jump = code[frame->f_lasti*sizeof(_Py_CODEUNIT)+1];
             frame->f_lasti += jump;
diff --git a/Python/ceval.c b/Python/ceval.c
index 9aaddd9..2c524ab 100644
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -2650,32 +2650,25 @@ handle_eval_breaker:
             }
             assert (gen_status == PYGEN_NEXT);
             assert (retval != NULL);
-            frame->f_state = FRAME_SUSPENDED;
-            _PyFrame_SetStackPointer(frame, stack_pointer);
-            TRACE_FUNCTION_EXIT();
-            DTRACE_FUNCTION_EXIT();
-            _Py_LeaveRecursiveCall(tstate);
-            /* Restore previous cframe and return. */
-            tstate->cframe = cframe.previous;
-            tstate->cframe->use_tracing = cframe.use_tracing;
-            assert(tstate->cframe->current_frame == frame->previous);
-            assert(!_PyErr_Occurred(tstate));
-            return retval;
+            PUSH(retval);
+            DISPATCH();
+        }
+
+        TARGET(ASYNC_GEN_WRAP) {
+            PyObject *v = TOP();
+            assert(frame->f_code->co_flags & CO_ASYNC_GENERATOR);
+            PyObject *w = _PyAsyncGenValueWrapperNew(v);
+            if (w == NULL) {
+                goto error;
+            }
+            SET_TOP(w);
+            Py_DECREF(v);
+            DISPATCH();
         }
 
         TARGET(YIELD_VALUE) {
             assert(frame->is_entry);
             PyObject *retval = POP();
-
-            if (frame->f_code->co_flags & CO_ASYNC_GENERATOR) {
-                PyObject *w = _PyAsyncGenValueWrapperNew(retval);
-                Py_DECREF(retval);
-                if (w == NULL) {
-                    retval = NULL;
-                    goto error;
-                }
-                retval = w;
-            }
             frame->f_state = FRAME_SUSPENDED;
             _PyFrame_SetStackPointer(frame, stack_pointer);
             TRACE_FUNCTION_EXIT();
diff --git a/Python/compile.c b/Python/compile.c
index 5d32959..feb9fca 100644
--- a/Python/compile.c
+++ b/Python/compile.c
@@ -910,6 +910,7 @@ stack_effect(int opcode, int oparg, int jump)
             return -1;
         case SETUP_ANNOTATIONS:
             return 0;
+        case ASYNC_GEN_WRAP:
         case YIELD_VALUE:
             return 0;
         case POP_BLOCK:
@@ -1541,6 +1542,9 @@ compiler_addop_j_noline(struct compiler *c, int opcode, basicblock *b)
 #define POP_EXCEPT_AND_RERAISE(C) \
     RETURN_IF_FALSE(compiler_pop_except_and_reraise((C)))
 
+#define ADDOP_YIELD(C) \
+    RETURN_IF_FALSE(addop_yield(C))
+
 #define VISIT(C, TYPE, V) {\
     if (!compiler_visit_ ## TYPE((C), (V))) \
         return 0; \
@@ -1844,6 +1848,7 @@ compiler_add_yield_from(struct compiler *c, int await)
     compiler_use_next_block(c, start);
     ADDOP_JUMP(c, SEND, exit);
     compiler_use_next_block(c, resume);
+    ADDOP(c, YIELD_VALUE);
     ADDOP_I(c, RESUME, await ? 3 : 2);
     ADDOP_JUMP(c, JUMP_NO_INTERRUPT, start);
     compiler_use_next_block(c, exit);
@@ -4094,6 +4099,17 @@ addop_binary(struct compiler *c, operator_ty binop, bool inplace)
     return 1;
 }
 
+
+static int
+addop_yield(struct compiler *c) {
+    if (c->u->u_ste->ste_generator && c->u->u_ste->ste_coroutine) {
+        ADDOP(c, ASYNC_GEN_WRAP);
+    }
+    ADDOP(c, YIELD_VALUE);
+    ADDOP_I(c, RESUME, 1);
+    return 1;
+}
+
 static int
 compiler_nameop(struct compiler *c, identifier name, expr_context_ty ctx)
 {
@@ -5144,8 +5160,7 @@ compiler_sync_comprehension_generator(struct compiler *c,
         switch (type) {
         case COMP_GENEXP:
             VISIT(c, expr, elt);
-            ADDOP(c, YIELD_VALUE);
-            ADDOP_I(c, RESUME, 1);
+            ADDOP_YIELD(c);
             ADDOP(c, POP_TOP);
             break;
         case COMP_LISTCOMP:
@@ -5243,8 +5258,7 @@ compiler_async_comprehension_generator(struct compiler *c,
         switch (type) {
         case COMP_GENEXP:
             VISIT(c, expr, elt);
-            ADDOP(c, YIELD_VALUE);
-            ADDOP_I(c, RESUME, 1);
+            ADDOP_YIELD(c);
             ADDOP(c, POP_TOP);
             break;
         case COMP_LISTCOMP:
@@ -5714,8 +5728,7 @@ compiler_visit_expr1(struct compiler *c, expr_ty e)
         else {
             ADDOP_LOAD_CONST(c, Py_None);
         }
-        ADDOP(c, YIELD_VALUE);
-        ADDOP_I(c, RESUME, 1);
+        ADDOP_YIELD(c);
         break;
     case YieldFrom_kind:
         if (c->u->u_ste->ste_type != FunctionBlock)
diff --git a/Python/opcode_targets.h b/Python/opcode_targets.h
index 11ac0e9..c19cd0e 100644
--- a/Python/opcode_targets.h
+++ b/Python/opcode_targets.h
@@ -86,7 +86,7 @@ static void *opcode_targets[256] = {
     &&TARGET_IMPORT_STAR,
     &&TARGET_SETUP_ANNOTATIONS,
     &&TARGET_YIELD_VALUE,
-    &&TARGET_STORE_ATTR_WITH_HINT,
+    &&TARGET_ASYNC_GEN_WRAP,
     &&TARGET_PREP_RERAISE_STAR,
     &&TARGET_POP_EXCEPT,
     &&TARGET_STORE_NAME,
@@ -130,7 +130,7 @@ static void *opcode_targets[256] = {
     &&TARGET_POP_JUMP_IF_NOT_NONE,
     &&TARGET_POP_JUMP_IF_NONE,
     &&TARGET_RAISE_VARARGS,
-    &&TARGET_LOAD_FAST__LOAD_FAST,
+    &&TARGET_STORE_ATTR_WITH_HINT,
     &&TARGET_MAKE_FUNCTION,
     &&TARGET_BUILD_SLICE,
     &&TARGET_JUMP_NO_INTERRUPT,
@@ -139,20 +139,20 @@ static void *opcode_targets[256] = {
     &&TARGET_LOAD_DEREF,
     &&TARGET_STORE_DEREF,
     &&TARGET_DELETE_DEREF,
+    &&TARGET_LOAD_FAST__LOAD_FAST,
     &&TARGET_STORE_FAST__LOAD_FAST,
-    &&TARGET_LOAD_FAST__LOAD_CONST,
     &&TARGET_CALL_FUNCTION_EX,
-    &&TARGET_LOAD_CONST__LOAD_FAST,
+    &&TARGET_LOAD_FAST__LOAD_CONST,
     &&TARGET_EXTENDED_ARG,
     &&TARGET_LIST_APPEND,
     &&TARGET_SET_ADD,
     &&TARGET_MAP_ADD,
     &&TARGET_LOAD_CLASSDEREF,
     &&TARGET_COPY_FREE_VARS,
-    &&TARGET_STORE_FAST__STORE_FAST,
+    &&TARGET_LOAD_CONST__LOAD_FAST,
     &&TARGET_RESUME,
     &&TARGET_MATCH_CLASS,
-    &&_unknown_opcode,
+    &&TARGET_STORE_FAST__STORE_FAST,
     &&_unknown_opcode,
     &&TARGET_FORMAT_VALUE,
     &&TARGET_BUILD_CONST_KEY_MAP,