added sync_fast_locals option to eval and exec

Author: blhsing

Include/internal/pycore_global_objects_fini_generated.h

@@ -803,6 +803,7 @@ struct _Py_global_strings {
         STRUCT_FOR_ID(sub_key)
         STRUCT_FOR_ID(subcalls)
         STRUCT_FOR_ID(symmetric_difference_update)
+        STRUCT_FOR_ID(sync_fast_locals)
         STRUCT_FOR_ID(tabsize)
         STRUCT_FOR_ID(tag)
         STRUCT_FOR_ID(take_bytes)

Include/internal/pycore_global_strings.h

@@ -955,6 +955,8 @@ eval as builtin_eval
     /
     globals: object = None
     locals: object = None
+    *
+    sync_fast_locals: bool = False
 
 Evaluate the given source in the context of globals and locals.
 
@@ -967,8 +969,8 @@ If only globals is given, locals defaults to it.
 
 static PyObject *
 builtin_eval_impl(PyObject *module, PyObject *source, PyObject *globals,
-                  PyObject *locals)
-/*[clinic end generated code: output=0a0824aa70093116 input=7c7bce5299a89062]*/
+                  PyObject *locals, int sync_fast_locals)
+/*[clinic end generated code: output=a573401639e51347 input=440105eb08930503]*/
 {
     PyThreadState *tstate = _PyThreadState_GET();
     PyObject *result = NULL, *source_copy;
@@ -1037,6 +1039,10 @@ builtin_eval_impl(PyObject *module, PyObject *source, PyObject *globals,
                 "code object passed to eval() may not contain free variables");
             goto error;
         }
+        if (!sync_fast_locals && ((PyCodeObject *)source)->co_flags & CO_OPTIMIZED) {
+            Py_DECREF(locals);
+            locals = NULL;
+        }
         result = PyEval_EvalCode(source, globals, locals);
     }
     else {
@@ -1078,6 +1084,7 @@ exec as builtin_exec
     locals: object = None
     *
     closure: object(c_default="NULL") = None
+    sync_fast_locals: bool = False
 
 Execute the given source in the context of globals and locals.
 
@@ -1092,8 +1099,8 @@ when source is a code object requiring exactly that many cellvars.
 
 static PyObject *
 builtin_exec_impl(PyObject *module, PyObject *source, PyObject *globals,
-                  PyObject *locals, PyObject *closure)
-/*[clinic end generated code: output=7579eb4e7646743d input=25e989b6d87a3a21]*/
+                  PyObject *locals, PyObject *closure, int sync_fast_locals)
+/*[clinic end generated code: output=ceab303bd7575dcf input=3a4103a242b26356]*/
 {
     PyThreadState *tstate = _PyThreadState_GET();
     PyObject *v;
@@ -1189,6 +1196,10 @@ builtin_exec_impl(PyObject *module, PyObject *source, PyObject *globals,
             goto error;
         }
 
+        if (!sync_fast_locals && ((PyCodeObject *)source)->co_flags & CO_OPTIMIZED) {
+            Py_DECREF(locals);
+            locals = NULL;
+        }
         if (!closure) {
             v = PyEval_EvalCode(source, globals, locals);
         } else {
@@ -1225,7 +1236,7 @@ builtin_exec_impl(PyObject *module, PyObject *source, PyObject *globals,
     if (v == NULL)
         goto error;
     Py_DECREF(globals);
-    Py_DECREF(locals);
+    Py_XDECREF(locals);
     Py_DECREF(v);
     Py_RETURN_NONE;
 

Include/internal/pycore_runtime_init_generated.h

@@ -1529,6 +1529,46 @@ typedef struct {
     _PyStackRef stack[1];
 } _PyEntryFrame;
 
+static int
+_PyEval_SyncMappingToFast(_PyInterpreterFrame *frame)
+{
+    PyObject *mapping = frame->f_locals;
+    PyCodeObject *co = _PyFrame_GetCode(frame);
+    PyObject *names = co->co_localsplusnames;
+
+    for (int i = 0; i < co->co_nlocalsplus; i++) {
+        PyObject *name = PyTuple_GET_ITEM(names, i);
+        PyObject *value = PyObject_GetItem(mapping, name);
+        if (value != NULL) {
+            frame->localsplus[i] = PyStackRef_FromPyObjectSteal(value);
+        }
+        else {
+            PyErr_Clear();
+        }
+    }
+    return 0;
+}
+
+static int
+_PyEval_SyncFastToMapping(_PyInterpreterFrame *frame)
+{
+    PyObject *mapping = frame->f_locals;
+    PyCodeObject *co = _PyFrame_GetCode(frame);
+    PyObject *names = co->co_localsplusnames;
+
+    for (int i = 0; i < co->co_nlocalsplus; i++) {
+        _PyStackRef sref = frame->localsplus[i];
+        if (!PyStackRef_IsNull(sref)) {
+            PyObject *name = PyTuple_GET_ITEM(names, i);
+            PyObject *obj = PyStackRef_AsPyObjectSteal(sref);
+            if (PyObject_SetItem(mapping, name, obj) < 0) {
+                PyErr_Clear();
+            }
+        }
+    }
+    return 0;
+}
+
 PyObject* _Py_HOT_FUNCTION DONT_SLP_VECTORIZE
 _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int throwflag)
 {
@@ -1591,6 +1631,15 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
     frame->previous = &entry.frame;
     tstate->current_frame = frame;
     entry.frame.localsplus[0] = PyStackRef_NULL;
+    PyCodeObject *co = _PyFrame_GetCode(frame);
+    if ((co->co_flags & CO_OPTIMIZED) && 
+        frame->f_locals != NULL && 
+        frame->f_locals != frame->f_globals) 
+    {
+        if (_PyEval_SyncMappingToFast(frame) < 0) {
+            goto early_exit;
+        }
+    }
 #ifdef _Py_TIER2
     if (tstate->current_executor != NULL) {
         entry.frame.localsplus[0] = PyStackRef_FromPyObjectNew(tstate->current_executor);
@@ -2377,6 +2426,15 @@ clear_gen_frame(PyThreadState *tstate, _PyInterpreterFrame * frame)
 void
 _PyEval_FrameClearAndPop(PyThreadState *tstate, _PyInterpreterFrame * frame)
 {
+    PyCodeObject *co = _PyFrame_GetCode(frame);
+    if ((co->co_flags & CO_OPTIMIZED) && 
+        frame->f_locals != NULL && 
+        frame->f_locals != frame->f_globals) 
+    {
+        if (_PyEval_SyncFastToMapping(frame) < 0) {
+            PyErr_WriteUnraisable(frame->f_locals);
+        }
+    }
     // Update last_profiled_frame for remote profiler frame caching.
     // By this point, tstate->current_frame is already set to the parent frame.
     // Only update if we're popping the exact frame that was last profiled.