some optimize

include/pocketpy/frame.h

@@ -90,8 +90,8 @@ struct Frame {
     NameDict& f_globals() noexcept { return _module->attr(); }
     PyObject* f_closure_try_get(StrName name);
 
-    Frame(PyObject** p0, const CodeObject* co, PyObject* _module, PyObject* _callable)
+    Frame(PyObject** p0, const CodeObject* co, PyObject* _module, PyObject* _callable, PyObject** _locals_base)
-            : _ip(-1), _next_ip(0), _sp_base(p0), co(co), _module(_module), _callable(_callable), _locals(co, p0) { }
+            : _ip(-1), _next_ip(0), _sp_base(p0), co(co), _module(_module), _callable(_callable), _locals(co, _locals_base) { }
 
     Frame(PyObject** p0, const CodeObject* co, PyObject* _module, PyObject* _callable, FastLocals _locals)
             : _ip(-1), _next_ip(0), _sp_base(p0), co(co), _module(_module), _callable(_callable), _locals(_locals) { }
@@ -150,6 +150,9 @@ struct CallStack{
     }
 
     void pop(){
+#if PK_DEBUG_EXTRA_CHECK
+        if(empty()) PK_FATAL_ERROR();
+#endif
         LinkedFrame* p = _tail;
         _tail = p->f_back;
         pool64_dealloc(p);

src/ceval.cpp

@@ -67,7 +67,7 @@ PyObject* VM::_run_top_frame(){
 __NEXT_FRAME:
     // cache
     const CodeObject* co = frame->co;
-    const auto& co_consts = co->consts;
+    PyObject** co_consts = const_cast<PyObject**>(co->consts.data());
     const Bytecode* co_codes = co->codes.data();
 
     Bytecode byte = co_codes[frame->next_bytecode()];

src/profiler.cpp

@@ -55,7 +55,7 @@ void LineProfiler::_step_end(LinkedFrame* linked_frame, int line){
     }else if(linked_frame->f_back == top_frame_record.frame){
         id_delta = 1;
     }else{
-        id_delta = -1;
+        id_delta = -1;  // unsafe
     }
 
     // current line is about to change

src/vm.cpp

@@ -930,7 +930,7 @@ PyObject* VM::vectorcall(int ARGC, int KWARGC, bool op_call){
         if(co->is_generator){
             s_data.reset(p0);
             return _py_generator(
-                Frame(nullptr, co, fn._module, callable),
+                Frame(nullptr, co, fn._module, callable, nullptr),
                 ArgsView(buffer, buffer + co_nlocals)
             );
         }
@@ -940,7 +940,7 @@ PyObject* VM::vectorcall(int ARGC, int KWARGC, bool op_call){
         for(int j=0; j<co_nlocals; j++) _base[j] = buffer[j];
 
 __FAST_CALL:
-        callstack.emplace(p0, co, fn._module, callable, FastLocals(co, args.begin()));
+        callstack.emplace(p0, co, fn._module, callable, args.begin());
         if(op_call) return PY_OP_CALL;
         return _run_top_frame();
         /*****************_py_call*****************/