gh-92777: Add LOAD_METHOD_LAZY_DICT (GH-92778)

3 files changed, 50 insertions, 18 deletions

diff --git a/Python/ceval.c b/Python/ceval.c
index e82f734..56b1017 100644
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -4673,6 +4673,29 @@ handle_eval_breaker:
             NOTRACE_DISPATCH();
         }
 
+        TARGET(LOAD_METHOD_LAZY_DICT) {
+            assert(cframe.use_tracing == 0);
+            PyObject *self = TOP();
+            PyTypeObject *self_cls = Py_TYPE(self);
+            _PyLoadMethodCache *cache = (_PyLoadMethodCache *)next_instr;
+            uint32_t type_version = read_u32(cache->type_version);
+            DEOPT_IF(self_cls->tp_version_tag != type_version, LOAD_METHOD);
+            int dictoffset = cache->dict_offset;
+            PyObject *dict = *(PyObject **)((char *)self + dictoffset);
+            assert(dictoffset == self_cls->tp_dictoffset && dictoffset > 0);
+            /* This object has a __dict__, just not yet created */
+            DEOPT_IF(dict != NULL, LOAD_METHOD);
+            STAT_INC(LOAD_METHOD, hit);
+            PyObject *res = read_obj(cache->descr);
+            assert(res != NULL);
+            assert(_PyType_HasFeature(Py_TYPE(res), Py_TPFLAGS_METHOD_DESCRIPTOR));
+            Py_INCREF(res);
+            SET_TOP(res);
+            PUSH(self);
+            JUMPBY(INLINE_CACHE_ENTRIES_LOAD_METHOD);
+            NOTRACE_DISPATCH();
+        }
+
         TARGET(LOAD_METHOD_MODULE) {
             /* LOAD_METHOD, for module methods */
             assert(cframe.use_tracing == 0);
diff --git a/Python/opcode_targets.h b/Python/opcode_targets.h
index 0d8faf2..71b7a76 100644
--- a/Python/opcode_targets.h
+++ b/Python/opcode_targets.h
@@ -80,12 +80,12 @@ static void *opcode_targets[256] = {
     &&TARGET_LOAD_GLOBAL_MODULE,
     &&TARGET_LOAD_METHOD_ADAPTIVE,
     &&TARGET_LOAD_METHOD_CLASS,
-    &&TARGET_LOAD_METHOD_MODULE,
+    &&TARGET_LOAD_METHOD_LAZY_DICT,
     &&TARGET_LIST_TO_TUPLE,
     &&TARGET_RETURN_VALUE,
     &&TARGET_IMPORT_STAR,
     &&TARGET_SETUP_ANNOTATIONS,
-    &&TARGET_LOAD_METHOD_NO_DICT,
+    &&TARGET_LOAD_METHOD_MODULE,
     &&TARGET_ASYNC_GEN_WRAP,
     &&TARGET_PREP_RERAISE_STAR,
     &&TARGET_POP_EXCEPT,
@@ -112,7 +112,7 @@ static void *opcode_targets[256] = {
     &&TARGET_JUMP_FORWARD,
     &&TARGET_JUMP_IF_FALSE_OR_POP,
     &&TARGET_JUMP_IF_TRUE_OR_POP,
-    &&TARGET_LOAD_METHOD_WITH_DICT,
+    &&TARGET_LOAD_METHOD_NO_DICT,
     &&TARGET_POP_JUMP_FORWARD_IF_FALSE,
     &&TARGET_POP_JUMP_FORWARD_IF_TRUE,
     &&TARGET_LOAD_GLOBAL,
@@ -120,13 +120,13 @@ static void *opcode_targets[256] = {
     &&TARGET_CONTAINS_OP,
     &&TARGET_RERAISE,
     &&TARGET_COPY,
-    &&TARGET_LOAD_METHOD_WITH_VALUES,
+    &&TARGET_LOAD_METHOD_WITH_DICT,
     &&TARGET_BINARY_OP,
     &&TARGET_SEND,
     &&TARGET_LOAD_FAST,
     &&TARGET_STORE_FAST,
     &&TARGET_DELETE_FAST,
-    &&TARGET_RESUME_QUICK,
+    &&TARGET_LOAD_METHOD_WITH_VALUES,
     &&TARGET_POP_JUMP_FORWARD_IF_NOT_NONE,
     &&TARGET_POP_JUMP_FORWARD_IF_NONE,
     &&TARGET_RAISE_VARARGS,
@@ -140,9 +140,9 @@ static void *opcode_targets[256] = {
     &&TARGET_STORE_DEREF,
     &&TARGET_DELETE_DEREF,
     &&TARGET_JUMP_BACKWARD,
-    &&TARGET_STORE_ATTR_ADAPTIVE,
+    &&TARGET_RESUME_QUICK,
     &&TARGET_CALL_FUNCTION_EX,
-    &&TARGET_STORE_ATTR_INSTANCE_VALUE,
+    &&TARGET_STORE_ATTR_ADAPTIVE,
     &&TARGET_EXTENDED_ARG,
     &&TARGET_LIST_APPEND,
     &&TARGET_SET_ADD,
@@ -152,30 +152,31 @@ static void *opcode_targets[256] = {
     &&TARGET_YIELD_VALUE,
     &&TARGET_RESUME,
     &&TARGET_MATCH_CLASS,
+    &&TARGET_STORE_ATTR_INSTANCE_VALUE,
     &&TARGET_STORE_ATTR_SLOT,
-    &&TARGET_STORE_ATTR_WITH_HINT,
     &&TARGET_FORMAT_VALUE,
     &&TARGET_BUILD_CONST_KEY_MAP,
     &&TARGET_BUILD_STRING,
+    &&TARGET_STORE_ATTR_WITH_HINT,
     &&TARGET_STORE_FAST__LOAD_FAST,
-    &&TARGET_STORE_FAST__STORE_FAST,
     &&TARGET_LOAD_METHOD,
-    &&TARGET_STORE_SUBSCR_ADAPTIVE,
+    &&TARGET_STORE_FAST__STORE_FAST,
     &&TARGET_LIST_EXTEND,
     &&TARGET_SET_UPDATE,
     &&TARGET_DICT_MERGE,
     &&TARGET_DICT_UPDATE,
+    &&TARGET_STORE_SUBSCR_ADAPTIVE,
     &&TARGET_STORE_SUBSCR_DICT,
     &&TARGET_STORE_SUBSCR_LIST_INT,
     &&TARGET_UNPACK_SEQUENCE_ADAPTIVE,
     &&TARGET_UNPACK_SEQUENCE_LIST,
-    &&TARGET_UNPACK_SEQUENCE_TUPLE,
     &&TARGET_CALL,
     &&TARGET_KW_NAMES,
     &&TARGET_POP_JUMP_BACKWARD_IF_NOT_NONE,
     &&TARGET_POP_JUMP_BACKWARD_IF_NONE,
     &&TARGET_POP_JUMP_BACKWARD_IF_FALSE,
     &&TARGET_POP_JUMP_BACKWARD_IF_TRUE,
+    &&TARGET_UNPACK_SEQUENCE_TUPLE,
     &&TARGET_UNPACK_SEQUENCE_TWO_TUPLE,
     &&_unknown_opcode,
     &&_unknown_opcode,
@@ -253,6 +254,5 @@ static void *opcode_targets[256] = {
     &&_unknown_opcode,
     &&_unknown_opcode,
     &&_unknown_opcode,
-    &&_unknown_opcode,
     &&TARGET_DO_TRACING
 };
diff --git a/Python/specialize.c b/Python/specialize.c
index 9579369..7148b5d 100644
--- a/Python/specialize.c
+++ b/Python/specialize.c
@@ -903,7 +903,8 @@ typedef enum {
     MANAGED_VALUES = 1,
     MANAGED_DICT = 2,
     OFFSET_DICT = 3,
-    NO_DICT = 4
+    NO_DICT = 4,
+    LAZY_DICT = 5,    
 } ObjectDictKind;
 
 // Please collect stats carefully before and after modifying. A subtle change
@@ -972,14 +973,17 @@ _Py_Specialize_LoadMethod(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name)
         else {
             PyObject *dict = *(PyObject **) ((char *)owner + dictoffset);
             if (dict == NULL) {
-                SPECIALIZATION_FAIL(LOAD_METHOD, SPEC_FAIL_NO_DICT);
-                goto fail;
+                // This object will have a dict if user access __dict__
+                dictkind = LAZY_DICT;
+                keys = NULL;
+            }
+            else {
+                keys = ((PyDictObject *)dict)->ma_keys;
+                dictkind = OFFSET_DICT;
             }
-            keys = ((PyDictObject *)dict)->ma_keys;
-            dictkind = OFFSET_DICT;
         }
     }
-    if (dictkind != NO_DICT) {
+    if (dictkind == MANAGED_VALUES || dictkind == MANAGED_DICT || dictkind == OFFSET_DICT) {
         Py_ssize_t index = _PyDictKeys_StringLookup(keys, name);
         if (index != DKIX_EMPTY) {
             SPECIALIZATION_FAIL(LOAD_METHOD, SPEC_FAIL_LOAD_METHOD_IS_ATTR);
@@ -1008,6 +1012,11 @@ _Py_Specialize_LoadMethod(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name)
             cache->dict_offset = (uint16_t)owner_cls->tp_dictoffset;
             _Py_SET_OPCODE(*instr, LOAD_METHOD_WITH_DICT);
             break;
+        case LAZY_DICT:
+            assert(owner_cls->tp_dictoffset > 0 && owner_cls->tp_dictoffset <= INT16_MAX);
+            cache->dict_offset = (uint16_t)owner_cls->tp_dictoffset;
+            _Py_SET_OPCODE(*instr, LOAD_METHOD_LAZY_DICT);
+            break;
     }
     /* `descr` is borrowed. This is safe for methods (even inherited ones from
     *  super classes!) as long as tp_version_tag is validated for two main reasons: