[WIP] gh-106608: Have _PyUOpExecutorObject use a variable-length array

Include/internal/pycore_uops.h

@@ -17,15 +17,26 @@ typedef struct {
 } _PyUOpInstruction;
 
 typedef struct {
+    PyObject_VAR_HEAD
     _PyExecutorObject base;
-    _PyUOpInstruction trace[_Py_UOP_MAX_TRACE_LENGTH];  // TODO: variable length
+    _PyUOpInstruction trace[1];
 } _PyUOpExecutorObject;
 
 _PyInterpreterFrame *_PyUopExecute(
     _PyExecutorObject *executor,
     _PyInterpreterFrame *frame,
     PyObject **stack_pointer);
 
+/* Cast argument to _PyUOpExecutorObject* type. */
+#define _PyUOpExecutor_CAST(op) \
+    _Py_CAST(_PyUOpExecutorObject*, (op))
+
+static inline Py_ssize_t PyUOpExecutor_GET_SIZE(PyObject *op) {
+    _PyUOpExecutorObject *executor = _PyUOpExecutor_CAST(op);
+    return Py_SIZE(executor);
+}
+#define PyUOpExecutor_GET_SIZE(op) PyUOpExecutor_GET_SIZE(_PyObject_CAST(op))
+
 #ifdef __cplusplus
 }
 #endif

Python/executor.c

@@ -31,6 +31,39 @@
 #define ENABLE_SPECIALIZATION 0
 
 
+// TODO: continue editing this from tupleobject.c to fit _PyUOpExecutorObject
+static _PyUOpExecutorObject *
+tuple_alloc(Py_ssize_t size)
+{
+    _PyUOpExecutorObject *op = maybe_freelist_pop(size);
+    if (op == NULL) {
+        /* Check for overflow */
+        if ((size_t)size > ((size_t)PY_SSIZE_T_MAX - (sizeof(PyTupleObject) -
+                    sizeof(PyObject *))) / sizeof(PyObject *)) {
+            return (_PyUOpExecutorObject *)PyErr_NoMemory();
+        }
+        op = PyObject_GC_NewVar(_PyUOpExecutorObject, &PyTuple_Type, size);
+        if (op == NULL)
+            return NULL;
+    }
+    return op;
+}
+
+PyObject *
+PyUOpExecutor_New(_PyUOpInstruction trace[], Py_ssize_t size)
+{
+    _PyUOpExecutorObject *op;
+    op = tuple_alloc(size);
+    if (op == NULL) {
+        return NULL;
+    }
+    for (Py_ssize_t i = 0; i < size; i++) {
+        op->trace[i] = trace[i];
+    }
+    _PyObject_GC_TRACK(op);
+    return (PyObject *) op;
+}
+
 _PyInterpreterFrame *
 _PyUopExecute(_PyExecutorObject *executor, _PyInterpreterFrame *frame, PyObject **stack_pointer)
 {

Python/optimizer.c

@@ -320,13 +320,7 @@ uop_name(int index) {
 static Py_ssize_t
 uop_len(_PyUOpExecutorObject *self)
 {
-    int count = 0;
-    for (; count < _Py_UOP_MAX_TRACE_LENGTH; count++) {
-        if (self->trace[count].opcode == 0) {
-            break;
-        }
-    }
-    return count;
+    return PyUOpExecutor_GET_SIZE(self);
 }
 
 static PyObject *
@@ -703,6 +697,10 @@ uop_optimize(
         return -1;
     }
     executor->base.execute = _PyUopExecute;
+    /*
+    This should replace PyObject_New and memcpy
+    _PyUOpExecutorObject *executor = PyUOpExecutor_New(trace, trace_length);
+    */
     memcpy(executor->trace, trace, trace_length * sizeof(_PyUOpInstruction));
         if (trace_length < _Py_UOP_MAX_TRACE_LENGTH) {
             executor->trace[trace_length].opcode = 0;  // Sentinel