Fix #197 (#227)

* init

* some optimize

* Update frame.h

* remove `LOAD_INTEGER`

* Update vm.cpp

* some optimize

* some fix

* Revert "remove `LOAD_INTEGER`"

This reverts commit c0b965aee2f64fbfae0b20f41d714688649d20cf.

* some fix

* Update expr.cpp

* some fix

* Update retype.yml

README.md

@@ -49,7 +49,7 @@ Some variables can be set to control the build process:
 + `PK_BUILD_SHARED_LIB` - Build the shared  (for C-APIs only)
 + `PK_ENABLE_OS` - Enable OS related features (default mode is sandboxed)
 
-It is safe to use `main` branch in production.
+It is safe to use `main` branch in production if CI is green.
 
 ### Compile Flags
 
@@ -197,10 +197,6 @@ Your sponsorship will help us develop pkpy continuously.
 
   An excellent learning material. It illustrates how Python's virtual machine works.
 
-+ [box2d](https://box2d.org/)
-
-  The world's best 2D physics engine, written by Erin Catto. `box2d` now becomes a built-in module in pkpy `v1.1.3` and later.
-
 
 ## Star History
 

docs/retype.yml

@@ -3,7 +3,7 @@ output: .retype
 url: https://pocketpy.dev
 branding:
   title: pocketpy
-  label: v1.4.1
+  label: v1.4.3
   logo: "./static/logo.png"
 favicon: "./static/logo.png"
 meta:

include/pocketpy/common.h

@@ -21,7 +21,7 @@
 #include <typeinfo>
 #include <initializer_list>
 
-#define PK_VERSION				"1.4.2"
+#define PK_VERSION				"1.4.3"
 
 #include "config.h"
 #include "export.h"

include/pocketpy/expr.h

@@ -105,6 +105,7 @@ struct CodeEmitContext{
     void exit_block();
     void emit_expr();   // clear the expression stack and generate bytecode
     int emit_(Opcode opcode, uint16_t arg, int line, bool is_virtual=false);
+    int emit_int(i64 value, int line);
     void patch_jump(int index);
     bool add_label(StrName name);
     int add_varname(StrName name);

include/pocketpy/frame.h

@@ -84,18 +84,21 @@ struct Frame {
 
     const CodeObject* co;
     PyObject* _module;
-    PyObject* _callable;    // weak ref
+    PyObject* _callable;    // a function object or nullptr (global scope)
     FastLocals _locals;
 
     NameDict& f_globals() noexcept { return _module->attr(); }
     PyObject* f_closure_try_get(StrName name);
 
-    Frame(PyObject** p0, const CodeObject* co, PyObject* _module, PyObject* _callable)
-            : _ip(-1), _next_ip(0), _sp_base(p0), co(co), _module(_module), _callable(_callable), _locals(co, p0) { }
+    // function scope
+    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, _locals_base) { }
 
+    // exec/eval
     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) { }
 
+    // global scope
     Frame(PyObject** p0, const CodeObject_& co, PyObject* _module)
             : _ip(-1), _next_ip(0), _sp_base(p0), co(co.get()), _module(_module), _callable(nullptr), _locals(co.get(), p0) {}
 
@@ -125,14 +128,46 @@ struct Frame {
     }
 };
 
-using CallstackContainer = small_vector_no_copy_and_move<Frame, 16>;
+struct LinkedFrame{
+    LinkedFrame* f_back;
+    Frame frame;
+    template<typename... Args>
+    LinkedFrame(LinkedFrame* f_back, Args&&... args) : f_back(f_back), frame(std::forward<Args>(args)...) {}
+};
+
+struct CallStack{
+    static_assert(sizeof(LinkedFrame) <= 64 && std::is_trivially_destructible_v<LinkedFrame>);
+
+    LinkedFrame* _tail;
+    int _size;
+    CallStack(): _tail(nullptr), _size(0) {}
+
+    int size() const { return _size; }
+    bool empty() const { return _size == 0; }
+    void clear(){ while(!empty()) pop(); }
+
+    template<typename... Args>
+    void emplace(Args&&... args){
+        _tail = new(pool64_alloc<LinkedFrame>()) LinkedFrame(_tail, std::forward<Args>(args)...);
+        ++_size;
+    }
 
-struct FrameId{
-    CallstackContainer* data;
-    int index;
-    FrameId(CallstackContainer* data, int index) : data(data), index(index) {}
-    Frame* operator->() const { return &data->operator[](index); }
-    Frame* get() const { return &data->operator[](index); }
+    void pop(){
+#if PK_DEBUG_EXTRA_CHECK
+        if(empty()) PK_FATAL_ERROR();
+#endif
+        LinkedFrame* p = _tail;
+        _tail = p->f_back;
+        pool64_dealloc(p);
+        --_size;
+    }
+
+    Frame& top() const { return _tail->frame; }
+
+    template<typename Func>
+    void apply(Func&& f){
+        for(LinkedFrame* p = _tail; p != nullptr; p = p->f_back) f(p->frame);
+    }
 };
 
 }; // namespace pkpy

include/pocketpy/profiler.h

@@ -14,7 +14,7 @@ struct _LineRecord{
 };
 
 struct _FrameRecord{
-    FrameId frame;
+    LinkedFrame* frame;
     clock_t prev_time;
     _LineRecord* prev_record;
 };
@@ -26,8 +26,8 @@ struct LineProfiler{
     std::set<FuncDecl*> functions;
 
     void begin();
-    void _step(FrameId frame);
-    void _step_end(FrameId frame, int line);
+    void _step(LinkedFrame*);
+    void _step_end(LinkedFrame*, int);
     void end();
     Str stats();
 };

include/pocketpy/vm.h

@@ -94,7 +94,7 @@ class VM {
 public:
     ManagedHeap heap;
     ValueStack s_data;
-    stack_no_copy<Frame, CallstackContainer> callstack;
+    CallStack callstack;
     std::vector<PyTypeInfo> _all_types;
     
     NameDict _modules;                                 // loaded modules
@@ -151,7 +151,7 @@ public:
 
     VM(bool enable_os=true);
 
-    FrameId top_frame();
+    Frame* top_frame();
     void _pop_frame();
 
     PyObject* py_str(PyObject* obj);
@@ -364,9 +364,9 @@ public:
     }
 
     Type _tp(PyObject* obj){
+        if(!is_tagged(obj)) return obj->type;
         if(is_int(obj)) return tp_int;
-        if(is_float(obj)) return tp_float;
-        return obj->type;
+        return tp_float;
     }
 
     PyObject* _t(PyObject* obj){

src/ceval.cpp

@@ -44,19 +44,12 @@ bool VM::py_ge(PyObject* _0, PyObject* _1){
 
 #undef BINARY_F_COMPARE
 
-// static i64 _py_sint(PyObject* obj) noexcept {
-//     return (i64)(PK_BITS(obj) >> 2);
-// }
-
 PyObject* VM::_run_top_frame(){
-    FrameId frame = top_frame();
-    const int base_id = frame.index;
+    Frame* frame = top_frame();
+    const Frame* base_frame = frame;
     bool need_raise = false;
 
     while(true){
-#if PK_DEBUG_EXTRA_CHECK
-        if(frame.index < base_id) PK_FATAL_ERROR();
-#endif
         try{
             if(need_raise){ need_raise = false; _raise(); }
 /**********************************************************************/
@@ -67,14 +60,14 @@ PyObject* VM::_run_top_frame(){
 {
 
 #define CEVAL_STEP_CALLBACK() \
-    if(_ceval_on_step) _ceval_on_step(this, frame.get(), byte); \
-    if(_profiler) _profiler->_step(frame);
+    if(_ceval_on_step) _ceval_on_step(this, frame, byte); \
+    if(_profiler) _profiler->_step(callstack._tail);
 
 #define DISPATCH_OP_CALL() { frame = top_frame(); goto __NEXT_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()];
@@ -598,7 +591,7 @@ __NEXT_STEP:;
     TARGET(RETURN_VALUE){
         PyObject* _0 = byte.arg == BC_NOARG ? POPX() : None;
         _pop_frame();
-        if(frame.index == base_id){       // [ frameBase<- ]
+        if(frame == base_frame){       // [ frameBase<- ]
             return _0;
         }else{
             frame = top_frame();
@@ -826,16 +819,12 @@ __NEXT_STEP:;
         }catch(UnhandledException){
             PyObject* e_obj = POPX();
             Exception& _e = PK_OBJ_GET(Exception, e_obj);
+            bool is_base_frame_to_be_popped = frame == base_frame;
             _pop_frame();
-            if(callstack.empty()){
-#if PK_DEBUG_FULL_EXCEPTION
-                std::cerr << _e.summary() << std::endl;
-#endif
-                throw _e;
-            }
+            if(callstack.empty()) throw _e;   // propagate to the top level
             frame = top_frame();
             PUSH(e_obj);
-            if(frame.index < base_id) throw ToBeRaisedException();
+            if(is_base_frame_to_be_popped) throw ToBeRaisedException();
             need_raise = true;
         }catch(ToBeRaisedException){
             need_raise = true;

src/expr.cpp

@@ -64,6 +64,14 @@ namespace pkpy{
         return i;
     }
 
+    int CodeEmitContext::emit_int(i64 _val, int line){
+        if(is_imm_int(_val)){
+            return emit_(OP_LOAD_INTEGER, (uint16_t)_val, line);
+        }else{
+            return emit_(OP_LOAD_CONST, add_const(VAR(_val)), line);
+        }
+    }
+
     void CodeEmitContext::patch_jump(int index) {
         int target = co->codes.size();
         co->codes[index].arg = target;
@@ -246,11 +254,7 @@ namespace pkpy{
         VM* vm = ctx->vm;
         if(std::holds_alternative<i64>(value)){
             i64 _val = std::get<i64>(value);
-            if(is_imm_int(_val)){
-                ctx->emit_(OP_LOAD_INTEGER, (uint16_t)_val, line);
-                return;
-            }
-            ctx->emit_(OP_LOAD_CONST, ctx->add_const(VAR(_val)), line);
+            ctx->emit_int(_val, line);
             return;
         }
         if(std::holds_alternative<f64>(value)){
@@ -272,11 +276,7 @@ namespace pkpy{
             LiteralExpr* lit = static_cast<LiteralExpr*>(child.get());
             if(std::holds_alternative<i64>(lit->value)){
                 i64 _val = -std::get<i64>(lit->value);
-                if(is_imm_int(_val)){
-                    ctx->emit_(OP_LOAD_INTEGER, (uint16_t)_val, line);
-                }else{
-                    ctx->emit_(OP_LOAD_CONST, ctx->add_const(VAR(_val)), line);
-                }
+                ctx->emit_int(_val, line);
                 return;
             }
             if(std::holds_alternative<f64>(lit->value)){
@@ -594,8 +594,8 @@ namespace pkpy{
             // vectorcall protocol
             for(auto& item: args) item->emit_(ctx);
             for(auto& item: kwargs){
-                uint16_t index = StrName(item.first.sv()).index;
-                ctx->emit_(OP_LOAD_INTEGER, index, line);
+                i64 _val = StrName(item.first.sv()).index;
+                ctx->emit_int(_val, line);
                 item.second->emit_(ctx);
             }
             int KWARGC = kwargs.size();

src/iter.cpp

@@ -48,7 +48,7 @@ namespace pkpy{
         // restore the context
         for(PyObject* obj: s_backup) vm->s_data.push(obj);
         s_backup.clear();
-        vm->callstack.push(std::move(frame));
+        vm->callstack.emplace(std::move(frame));
 
         PyObject* ret;
         try{

src/pocketpy.cpp

@@ -77,7 +77,7 @@ void init_builtins(VM* _vm) {
             class_arg = args[0];
             self_arg = args[1];
         }else if(args.size() == 0){
-            FrameId frame = vm->top_frame();
+            Frame* frame = vm->top_frame();
             if(frame->_callable != nullptr){
                 class_arg = PK_OBJ_GET(Function, frame->_callable)._class;
                 if(frame->_locals.size() > 0) self_arg = frame->_locals[0];
@@ -184,7 +184,7 @@ void init_builtins(VM* _vm) {
         CodeObject_ code = vm->compile(CAST(Str&, args[0]), "<eval>", EVAL_MODE, true);
         PyObject* globals = args[1];
         if(globals == vm->None){
-            FrameId frame = vm->top_frame();
+            Frame* frame = vm->top_frame();
             return vm->_exec(code.get(), frame->_module, frame->_callable, frame->_locals);
         }
         vm->check_non_tagged_type(globals, vm->tp_mappingproxy);
@@ -196,7 +196,7 @@ void init_builtins(VM* _vm) {
         CodeObject_ code = vm->compile(CAST(Str&, args[0]), "<exec>", EXEC_MODE, true);
         PyObject* globals = args[1];
         if(globals == vm->None){
-            FrameId frame = vm->top_frame();
+            Frame* frame = vm->top_frame();
             vm->_exec(code.get(), frame->_module, frame->_callable, frame->_locals);
             return vm->None;
         }

src/profiler.cpp

@@ -18,16 +18,17 @@ void LineProfiler::begin(){
     frames.clear();
 }
 
-void LineProfiler::_step(FrameId frame){
+void LineProfiler::_step(LinkedFrame* linked_frame){
+    Frame* frame = &linked_frame->frame;
     auto line_info = frame->co->lines[frame->_ip];
     if(line_info.is_virtual) return;
     std::string_view filename = frame->co->src->filename.sv();
     int line = line_info.lineno;
 
     if(frames.empty()){
-        frames.push({frame, clock(), nullptr});
+        frames.push({linked_frame, clock(), nullptr});
     }else{
-        _step_end(frame, line);
+        _step_end(linked_frame, line);
     }
 
     auto& file_records = records[filename];
@@ -43,13 +44,19 @@ void LineProfiler::_step(FrameId frame){
     frames.top().prev_record = &file_records.at(line);
 }
 
-void LineProfiler::_step_end(FrameId frame, int line){
+void LineProfiler::_step_end(LinkedFrame* linked_frame, int line){
     clock_t now = clock();
     _FrameRecord& top_frame_record = frames.top();
     _LineRecord* prev_record = top_frame_record.prev_record;
 
-    int id_delta = frame.index - top_frame_record.frame.index;
-    PK_ASSERT(id_delta >= -1 && id_delta <= 1);
+    int id_delta;
+    if(linked_frame == top_frame_record.frame){
+        id_delta = 0;
+    }else if(linked_frame->f_back == top_frame_record.frame){
+        id_delta = 1;
+    }else{
+        id_delta = -1;  // unsafe
+    }
 
     // current line is about to change
     if(prev_record->line != line){
@@ -60,7 +67,7 @@ void LineProfiler::_step_end(FrameId frame, int line){
     }
     
     if(id_delta == 1){
-        frames.push({frame, now, nullptr});
+        frames.push({linked_frame, now, nullptr});
     }else{
         if(id_delta == -1) frames.pop();
     }

src/vm.cpp

@@ -119,16 +119,15 @@ namespace pkpy{
         PK_UNREACHABLE();
     }
 
-    FrameId VM::top_frame(){
+    Frame* VM::top_frame(){
 #if PK_DEBUG_EXTRA_CHECK
         if(callstack.empty()) PK_FATAL_ERROR();
 #endif
-        return FrameId(&callstack.container(), callstack.size()-1);
+        return &callstack.top();
     }
 
     void VM::_pop_frame(){
-        Frame* frame = &callstack.top();
-        s_data.reset(frame->_sp_base);
+        s_data.reset(callstack.top()._sp_base);
         callstack.pop();
     }
 
@@ -652,7 +651,7 @@ void VM::_log_s_data(const char* title) {
         if(f._sp_base == nullptr) PK_FATAL_ERROR();
         sp_bases[f._sp_base] += 1;
     }
-    FrameId frame = top_frame();
+    Frame* frame = top_frame();
     int line = frame->co->lines[frame->_ip];
     ss << frame->co->name << ":" << line << " [";
     for(PyObject** p=s_data.begin(); p!=s_data.end(); p++){
@@ -838,7 +837,7 @@ void VM::_prepare_py_call(PyObject** buffer, ArgsView args, ArgsView kwargs, con
     }
 
     for(int j=0; j<kwargs.size(); j+=2){
-        StrName key(CAST(int, kwargs[j]));
+        StrName key(_CAST(uint16_t, kwargs[j]));
         int index = decl->kw_to_index.try_get_likely_found(key);
         // if key is an explicit key, set as local variable
         if(index >= 0){
@@ -861,13 +860,16 @@ PyObject* VM::vectorcall(int ARGC, int KWARGC, bool op_call){
     // [callable, <self>, args..., kwargs...]
     //      ^p0                    ^p1      ^_sp
     PyObject* callable = p1[-(ARGC + 2)];
+    Type callable_t = _tp(callable);
+
     bool method_call = p1[-(ARGC + 1)] != PY_NULL;
 
     // handle boundmethod, do a patch
-    if(is_non_tagged_type(callable, tp_bound_method)){
+    if(callable_t == tp_bound_method){
         if(method_call) PK_FATAL_ERROR();
         BoundMethod& bm = PK_OBJ_GET(BoundMethod, callable);
         callable = bm.func;      // get unbound method
+        callable_t = _tp(callable);
         p1[-(ARGC + 2)] = bm.func;
         p1[-(ARGC + 1)] = bm.self;
         method_call = true;
@@ -880,26 +882,7 @@ PyObject* VM::vectorcall(int ARGC, int KWARGC, bool op_call){
     PyObject** _base = args.begin();
     PyObject* buffer[PK_MAX_CO_VARNAMES];
 
-    if(is_non_tagged_type(callable, tp_native_func)){
-        const auto& f = PK_OBJ_GET(NativeFunc, callable);
-        PyObject* ret;
-        if(f.decl != nullptr){
-            int co_nlocals = f.decl->code->varnames.size();
-            _prepare_py_call(buffer, args, kwargs, f.decl);
-            // copy buffer back to stack
-            s_data.reset(_base + co_nlocals);
-            for(int j=0; j<co_nlocals; j++) _base[j] = buffer[j];
-            ret = f.call(vm, ArgsView(s_data._sp - co_nlocals, s_data._sp));
-        }else{
-            if(KWARGC != 0) TypeError("old-style native_func does not accept keyword arguments");
-            f.check_size(this, args);
-            ret = f.call(this, args);
-        }
-        s_data.reset(p0);
-        return ret;
-    }
-
-    if(is_non_tagged_type(callable, tp_function)){
+    if(callable_t == tp_function){
         /*****************_py_call*****************/
         // callable must be a `function` object
         if(s_data.is_overflow()) StackOverflowError();
@@ -931,7 +914,7 @@ 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)
             );
         }
@@ -941,13 +924,32 @@ 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*****************/
     }
 
-    if(is_non_tagged_type(callable, tp_type)){
+    if(callable_t == tp_native_func){
+        const auto& f = PK_OBJ_GET(NativeFunc, callable);
+        PyObject* ret;
+        if(f.decl != nullptr){
+            int co_nlocals = f.decl->code->varnames.size();
+            _prepare_py_call(buffer, args, kwargs, f.decl);
+            // copy buffer back to stack
+            s_data.reset(_base + co_nlocals);
+            for(int j=0; j<co_nlocals; j++) _base[j] = buffer[j];
+            ret = f.call(vm, ArgsView(s_data._sp - co_nlocals, s_data._sp));
+        }else{
+            if(KWARGC != 0) TypeError("old-style native_func does not accept keyword arguments");
+            f.check_size(this, args);
+            ret = f.call(this, args);
+        }
+        s_data.reset(p0);
+        return ret;
+    }
+
+    if(callable_t == tp_type){
         // [type, NULL, args..., kwargs...]
         PyObject* new_f = find_name_in_mro(PK_OBJ_GET(Type, callable), __new__);
         PyObject* obj;
@@ -970,6 +972,7 @@ __FAST_CALL:
         // __init__
         PyObject* self;
         callable = get_unbound_method(obj, __init__, &self, false);
+        callable_t = _tp(callable);
         if (self != PY_NULL) {
             // replace `NULL` with `self`
             p1[-(ARGC + 2)] = callable;
@@ -994,7 +997,7 @@ __FAST_CALL:
         // [call_f, self, args..., kwargs...]
         return vectorcall(ARGC, KWARGC, false);
     }
-    TypeError(_type_name(vm, _tp(callable)).escape() + " object is not callable");
+    TypeError(_type_name(vm, callable_t).escape() + " object is not callable");
     PK_UNREACHABLE()
 }
 
@@ -1238,7 +1241,7 @@ void VM::_error(PyObject* e_obj){
 }
 
 void VM::_raise(bool re_raise){
-    Frame* frame = top_frame().get();
+    Frame* frame = top_frame();
     Exception& e = PK_OBJ_GET(Exception, s_data.top());
     if(!re_raise){
         e._ip_on_error = frame->_ip;
@@ -1259,7 +1262,7 @@ void VM::_raise(bool re_raise){
 
 void ManagedHeap::mark() {
     for(PyObject* obj: _no_gc) PK_OBJ_MARK(obj);
-    for(auto& frame : vm->callstack.container()) frame._gc_mark();
+    vm->callstack.apply([](Frame& frame){ frame._gc_mark(); });
     for(PyObject* obj: vm->s_data) PK_OBJ_MARK(obj);
     for(auto [_, co]: vm->_cached_codes) co->_gc_mark();
     if(vm->_last_exception) PK_OBJ_MARK(vm->_last_exception);