gh-93911: Specialize LOAD_ATTR for custom __getattr__ and __getattribute__

Include/internal/pycore_code.h

@@ -118,6 +118,7 @@ struct callable_cache {
     PyObject *isinstance;
     PyObject *len;
     PyObject *list_append;
+    PyObject *object__getattribute__;
 };
 
 /* "Locals plus" for a code object is the set of locals + cell vars +

Include/internal/pycore_typeobject.h

@@ -75,6 +75,9 @@ extern void _PyStaticType_ClearWeakRefs(PyTypeObject *type);
 extern void _PyStaticType_Dealloc(PyTypeObject *type);
 
 
+PyObject *_Py_slot_tp_getattro(PyObject *self, PyObject *name);
+PyObject *_Py_slot_tp_getattr_hook(PyObject *self, PyObject *name);
+
 #ifdef __cplusplus
 }
 #endif

Lib/opcode.py

@@ -338,6 +338,7 @@ def pseudo_op(name, op, real_ops):
         "LOAD_ATTR_ADAPTIVE",
         # These potentially push [NULL, bound method] onto the stack.
         "LOAD_ATTR_CLASS",
+        "LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN",
         "LOAD_ATTR_INSTANCE_VALUE",
         "LOAD_ATTR_MODULE",
         "LOAD_ATTR_PROPERTY",

Misc/NEWS.d/next/Core and Builtins/2022-06-18-17-00-33.gh-issue-93911.y286of.rst

@@ -0,0 +1 @@
+Specialize ``LOAD_ATTR`` for objects with custom ``__getattr__`` and ``__getattribute__``.

Objects/typeobject.c

@@ -8053,17 +8053,17 @@ slot_tp_call(PyObject *self, PyObject *args, PyObject *kwds)
 
 /* There are two slot dispatch functions for tp_getattro.
 
-   - slot_tp_getattro() is used when __getattribute__ is overridden
+   - _Py_slot_tp_getattro() is used when __getattribute__ is overridden
      but no __getattr__ hook is present;
 
-   - slot_tp_getattr_hook() is used when a __getattr__ hook is present.
+   - _Py_slot_tp_getattr_hook() is used when a __getattr__ hook is present.
 
-   The code in update_one_slot() always installs slot_tp_getattr_hook(); this
-   detects the absence of __getattr__ and then installs the simpler slot if
-   necessary. */
+   The code in update_one_slot() always installs _Py_slot_tp_getattr_hook();
+   this detects the absence of __getattr__ and then installs the simpler
+   slot if necessary. */
 
-static PyObject *
-slot_tp_getattro(PyObject *self, PyObject *name)
+PyObject *
+_Py_slot_tp_getattro(PyObject *self, PyObject *name)
 {
     PyObject *stack[2] = {self, name};
     return vectorcall_method(&_Py_ID(__getattribute__), stack, 2);
@@ -8094,8 +8094,8 @@ call_attribute(PyObject *self, PyObject *attr, PyObject *name)
     return res;
 }
 
-static PyObject *
-slot_tp_getattr_hook(PyObject *self, PyObject *name)
+PyObject *
+_Py_slot_tp_getattr_hook(PyObject *self, PyObject *name)
 {
     PyTypeObject *tp = Py_TYPE(self);
     PyObject *getattr, *getattribute, *res;
@@ -8108,8 +8108,8 @@ slot_tp_getattr_hook(PyObject *self, PyObject *name)
     getattr = _PyType_Lookup(tp, &_Py_ID(__getattr__));
     if (getattr == NULL) {
         /* No __getattr__ hook: use a simpler dispatcher */
-        tp->tp_getattro = slot_tp_getattro;
-        return slot_tp_getattro(self, name);
+        tp->tp_getattro = _Py_slot_tp_getattro;
+        return _Py_slot_tp_getattro(self, name);
     }
     Py_INCREF(getattr);
     /* speed hack: we could use lookup_maybe, but that would resolve the
@@ -8469,10 +8469,10 @@ static slotdef slotdefs[] = {
            PyWrapperFlag_KEYWORDS),
     TPSLOT("__str__", tp_str, slot_tp_str, wrap_unaryfunc,
            "__str__($self, /)\n--\n\nReturn str(self)."),
-    TPSLOT("__getattribute__", tp_getattro, slot_tp_getattr_hook,
+    TPSLOT("__getattribute__", tp_getattro, _Py_slot_tp_getattr_hook,
            wrap_binaryfunc,
            "__getattribute__($self, name, /)\n--\n\nReturn getattr(self, name)."),
-    TPSLOT("__getattr__", tp_getattro, slot_tp_getattr_hook, NULL, ""),
+    TPSLOT("__getattr__", tp_getattro, _Py_slot_tp_getattr_hook, NULL, ""),
     TPSLOT("__setattr__", tp_setattro, slot_tp_setattro, wrap_setattr,
            "__setattr__($self, name, value, /)\n--\n\nImplement setattr(self, name, value)."),
     TPSLOT("__delattr__", tp_setattro, slot_tp_setattro, wrap_delattr,

Python/ceval.c

@@ -3698,6 +3698,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
             DEOPT_IF(cls->tp_version_tag != type_version, LOAD_ATTR);
             assert(type_version != 0);
             PyObject *fget = read_obj(cache->descr);
+            assert(Py_IS_TYPE(fget, &PyFunction_Type));
             PyFunctionObject *f = (PyFunctionObject *)fget;
             uint32_t func_version = read_u32(cache->keys_version);
             assert(func_version != 0);
@@ -3709,8 +3710,8 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
             Py_INCREF(fget);
             _PyInterpreterFrame *new_frame = _PyFrame_PushUnchecked(tstate, f);
             SET_TOP(NULL);
-            int push_null = !(oparg & 1);
-            STACK_SHRINK(push_null);
+            int shrink_stack = !(oparg & 1);
+            STACK_SHRINK(shrink_stack);
             new_frame->localsplus[0] = owner;
             for (int i = 1; i < code->co_nlocalsplus; i++) {
                 new_frame->localsplus[i] = NULL;
@@ -3724,6 +3725,44 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
             goto start_frame;
         }
 
+        TARGET(LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN) {
+            assert(cframe.use_tracing == 0);
+            DEOPT_IF(tstate->interp->eval_frame, LOAD_ATTR);
+            _PyLoadMethodCache *cache = (_PyLoadMethodCache *)next_instr;
+            PyObject *owner = TOP();
+            PyTypeObject *cls = Py_TYPE(owner);
+            uint32_t type_version = read_u32(cache->type_version);
+            DEOPT_IF(cls->tp_version_tag != type_version, LOAD_ATTR);
+            assert(type_version != 0);
+            PyObject *getattribute = read_obj(cache->descr);
+            assert(Py_IS_TYPE(getattribute, &PyFunction_Type));
+            PyFunctionObject *f = (PyFunctionObject *)getattribute;
+            PyCodeObject *code = (PyCodeObject *)f->func_code;
+            DEOPT_IF(((PyCodeObject *)f->func_code)->co_argcount != 2, LOAD_ATTR);
+            DEOPT_IF(!_PyThreadState_HasStackSpace(tstate, code->co_framesize), CALL);
+            STAT_INC(LOAD_ATTR, hit);
+
+            PyObject *name = GETITEM(names, oparg >> 1);
+            Py_INCREF(f);
+            _PyInterpreterFrame *new_frame = _PyFrame_PushUnchecked(tstate, f);
+            SET_TOP(NULL);
+            int shrink_stack = !(oparg & 1);
+            STACK_SHRINK(shrink_stack);
+            Py_INCREF(name);
+            new_frame->localsplus[0] = owner;
+            new_frame->localsplus[1] = name;
+            for (int i = 2; i < code->co_nlocalsplus; i++) {
+                new_frame->localsplus[i] = NULL;
+            }
+            _PyFrame_SetStackPointer(frame, stack_pointer);
+            JUMPBY(INLINE_CACHE_ENTRIES_LOAD_ATTR);
+            frame->prev_instr = next_instr - 1;
+            new_frame->previous = frame;
+            frame = cframe.current_frame = new_frame;
+            CALL_STAT_INC(inlined_py_calls);
+            goto start_frame;
+        }
+
         TARGET(STORE_ATTR_ADAPTIVE) {
             assert(cframe.use_tracing == 0);
             _PyAttrCache *cache = (_PyAttrCache *)next_instr;