Aggressive Parser/Compiler Removal for Bytecode-Only Runtime

Makefile

@@ -216,9 +216,9 @@ else
 QJSC_CC=$(CC)
 QJSC=./qjsc$(EXE)
 endif
-PROGS+=libquickjs.a
+PROGS+=libquickjs.a libquickjs-bytecode.a
 ifdef CONFIG_LTO
-PROGS+=libquickjs.lto.a
+PROGS+=libquickjs.lto.a libquickjs-bytecode.lto.a
 endif
 
 # examples
@@ -244,6 +244,7 @@ endif
 all: $(OBJDIR) $(OBJDIR)/quickjs.check.o $(OBJDIR)/qjs.check.o $(PROGS)
 
 QJS_LIB_OBJS=$(OBJDIR)/quickjs.o $(OBJDIR)/dtoa.o $(OBJDIR)/libregexp.o $(OBJDIR)/libunicode.o $(OBJDIR)/cutils.o $(OBJDIR)/quickjs-libc.o
+QJS_LIB_BYTECODE_OBJS=$(patsubst $(OBJDIR)/%.o, $(OBJDIR)/%.bytecode.o, $(QJS_LIB_OBJS))
 
 QJS_OBJS=$(OBJDIR)/qjs.o $(OBJDIR)/repl.o $(QJS_LIB_OBJS)
 
@@ -303,9 +304,15 @@ endif
 libquickjs$(LTOEXT).a: $(QJS_LIB_OBJS)
 	$(AR) rcs $@ $^
 
+libquickjs-bytecode$(LTOEXT).a: $(QJS_LIB_BYTECODE_OBJS)
+	$(AR) rcs $@ $^
+
 ifdef CONFIG_LTO
 libquickjs.a: $(patsubst %.o, %.nolto.o, $(QJS_LIB_OBJS))
 	$(AR) rcs $@ $^
+
+libquickjs-bytecode.a: $(patsubst %.o, %.bytecode.nolto.o, $(QJS_LIB_OBJS))
+	$(AR) rcs $@ $^
 endif # CONFIG_LTO
 
 libquickjs.fuzz.a: $(patsubst %.o, %.fuzz.o, $(QJS_LIB_OBJS))
@@ -344,6 +351,12 @@ $(OBJDIR)/%.pic.o: %.c | $(OBJDIR)
 $(OBJDIR)/%.nolto.o: %.c | $(OBJDIR)
 	$(CC) $(CFLAGS_NOLTO) -c -o $@ $<
 
+$(OBJDIR)/%.bytecode.o: %.c | $(OBJDIR)
+	$(CC) $(CFLAGS_OPT) -DCONFIG_BYTECODE_ONLY_RUNTIME -c -o $@ $<
+
+$(OBJDIR)/%.bytecode.nolto.o: %.c | $(OBJDIR)
+	$(CC) $(CFLAGS_NOLTO) -DCONFIG_BYTECODE_ONLY_RUNTIME -c -o $@ $<
+
 $(OBJDIR)/%.debug.o: %.c | $(OBJDIR)
 	$(CC) $(CFLAGS_DEBUG) -c -o $@ $<
 
@@ -389,6 +402,8 @@ HELLO_OPTS=-fno-string-normalize -fno-map -fno-promise -fno-typedarray \
            -fno-typedarray -fno-regexp -fno-json -fno-eval -fno-proxy \
            -fno-date -fno-module-loader
 
+BYTECODE_RUNTIME_OPTS=-fno-eval -fno-regexp -fno-json -fno-module-loader
+
 hello.c: $(QJSC) $(HELLO_SRCS)
 	$(QJSC) -e $(HELLO_OPTS) -o $@ $(HELLO_SRCS)
 
@@ -465,6 +480,27 @@ endif
 stats: qjs$(EXE)
 	$(WINE) ./qjs$(EXE) -qd
 
+test-bytecode-runtime: qjsc$(EXE) libquickjs-bytecode.lto.a
+	./qjsc -e $(BYTECODE_RUNTIME_OPTS) -o hello_bytecode.c examples/hello.js
+	$(CC) $(CFLAGS_OPT) -flto -o hello_bytecode hello_bytecode.c libquickjs-bytecode.lto.a $(LIBS)
+	./hello_bytecode | grep "Hello World"
+	./qjsc -e -m $(BYTECODE_RUNTIME_OPTS) -o test_bytecode_runtime.c tests/test_bytecode_runtime.js
+	$(CC) $(CFLAGS_OPT) -flto -o test_bytecode_runtime test_bytecode_runtime.c libquickjs-bytecode.lto.a $(LIBS)
+	./test_bytecode_runtime
+	@if nm hello_bytecode | grep -E "__JS_EvalInternal|js_parse_program|JS_ParseJSON3|js_compile_regexp|JS_LoadModule"; then \
+		echo "Error: forbidden symbols found in hello_bytecode"; \
+		nm hello_bytecode | grep -E "__JS_EvalInternal|js_parse_program|JS_ParseJSON3|js_compile_regexp|JS_LoadModule"; \
+		exit 1; \
+	fi
+	@echo "Testing corrupted bytecode rejection..."
+	echo "corrupted" > /tmp/bad.bin
+	@if ./hello_bytecode /tmp/bad.bin 2>&1 | grep -q "Segmentation fault"; then \
+		echo "Error: Bytecode-only runtime crashed on corrupted input"; \
+		exit 1; \
+	fi
+	@echo "Bytecode-only runtime test passed"
+	rm -f hello_bytecode hello_bytecode.c test_bytecode_runtime test_bytecode_runtime.c /tmp/bad.bin
+
 microbench: qjs$(EXE)
 	$(WINE) ./qjs$(EXE) --std tests/microbench.js
 

qjs.c

@@ -482,7 +482,7 @@ int main(int argc, char **argv)
     }
 
     if (!empty_run) {
-        js_std_add_helpers(ctx, argc - optind, argv + optind);
+        js_std_add_helpers(ctx, argc - optind, argv + optind, TRUE);
 
         /* make 'std' and 'os' visible to non module code */
         if (load_std) {

qjsc.c

@@ -36,6 +36,10 @@
 #include "cutils.h"
 #include "quickjs-libc.h"
 
+#ifdef CONFIG_BYTECODE_ONLY_RUNTIME
+#error "qjsc must be built with the full QuickJS engine"
+#endif
+
 typedef struct {
     char *name;
     char *short_name;
@@ -79,6 +83,15 @@ static const FeatureEntry feature_list[] = {
     { "weakref", "WeakRef" },
 };
 
+#define FE_MASK(i) ((uint64_t)1 << (i))
+#define BYTECODE_ONLY_TRIGGER_MASK \
+    (FE_MASK(1) | FE_MASK(3) | FE_MASK(4) | FE_MASK(FE_MODULE_LOADER))
+
+static BOOL runtime_needs_parser(void)
+{
+    return (feature_bitmap & BYTECODE_ONLY_TRIGGER_MASK) != 0;
+}
+
 void namelist_add(namelist_t *lp, const char *name, const char *short_name,
                   int flags)
 {
@@ -477,6 +490,8 @@ static int output_executable(const char *out_filename, const char *cfilename,
 
     lto_suffix = "";
     bn_suffix = "";
+    if (!runtime_needs_parser())
+        bn_suffix = "-bytecode";
 
     arg = argv;
     *arg++ = CONFIG_CC;
@@ -841,14 +856,10 @@ int main(int argc, char **argv)
                     (unsigned int)stack_size);
         }
 
-        /* add the module loader if necessary */
-        if (feature_bitmap & (1 << FE_MODULE_LOADER)) {
-            fprintf(fo, "  JS_SetModuleLoaderFunc2(rt, NULL, js_module_loader, js_module_check_attributes, NULL);\n");
-        }
-
         fprintf(fo,
                 "  ctx = JS_NewCustomContext(rt);\n"
-                "  js_std_add_helpers(ctx, argc, argv);\n");
+                "  js_std_add_helpers(ctx, argc, argv, %d);\n",
+                (feature_bitmap & (1 << FE_MODULE_LOADER)) != 0);
 
         for(i = 0; i < cname_list.count; i++) {
             namelist_entry_t *e = &cname_list.array[i];

quickjs-libc.c

@@ -67,7 +67,7 @@ typedef sig_t sighandler_t;
 /* enable the os.Worker API. It relies on POSIX threads */
 #define USE_WORKER
 
-#ifdef USE_WORKER
+#if defined(USE_WORKER) && !defined(CONFIG_BYTECODE_ONLY_RUNTIME)
 #include <pthread.h>
 #include <stdatomic.h>
 #endif
@@ -123,7 +123,7 @@ typedef struct JSWaker {
 
 typedef struct {
     int ref_count;
-#ifdef USE_WORKER
+#if defined(USE_WORKER) && !defined(CONFIG_BYTECODE_ONLY_RUNTIME)
     pthread_mutex_t mutex;
 #endif
     struct list_head msg_queue; /* list of JSWorkerMessage.link */
@@ -431,6 +431,7 @@ uint8_t *js_load_file(JSContext *ctx, size_t *pbuf_len, const char *filename)
 }
 
 /* load and evaluate a file */
+#ifndef CONFIG_BYTECODE_ONLY_RUNTIME
 static JSValue js_loadScript(JSContext *ctx, JSValueConst this_val,
                              int argc, JSValueConst *argv)
 {
@@ -454,6 +455,7 @@ static JSValue js_loadScript(JSContext *ctx, JSValueConst this_val,
     JS_FreeCString(ctx, filename);
     return ret;
 }
+#endif
 
 /* load a file as a UTF-8 encoded string */
 static JSValue js_std_loadFile(JSContext *ctx, JSValueConst this_val,
@@ -679,11 +681,16 @@ JSModuleDef *js_module_loader(JSContext *ctx,
                               JSValueConst attributes)
 {
     JSModuleDef *m;
-    int res;
 
     if (has_suffix(module_name, ".so")) {
         m = js_module_loader_so(ctx, module_name);
     } else {
+#ifdef CONFIG_BYTECODE_ONLY_RUNTIME
+        JS_ThrowTypeError(ctx, "could not load module filename '%s': source module loading is not supported",
+                          module_name);
+        return NULL;
+#else
+        int res;
         size_t buf_len;
         uint8_t *buf;
 
@@ -723,6 +730,7 @@ JSModuleDef *js_module_loader(JSContext *ctx,
             m = JS_VALUE_GET_PTR(func_val);
             JS_FreeValue(ctx, func_val);
         }
+#endif
     }
     return m;
 }
@@ -870,6 +878,7 @@ static int get_bool_option(JSContext *ctx, BOOL *pbool,
     return 0;
 }
 
+#ifndef CONFIG_BYTECODE_ONLY_RUNTIME
 static JSValue js_evalScript(JSContext *ctx, JSValueConst this_val,
                              int argc, JSValueConst *argv)
 {
@@ -918,6 +927,7 @@ static JSValue js_evalScript(JSContext *ctx, JSValueConst this_val,
     }
     return ret;
 }
+#endif
 
 static JSClassID js_std_file_class_id;
 
@@ -957,6 +967,7 @@ static JSValue js_std_strerror(JSContext *ctx, JSValueConst this_val,
     return JS_NewString(ctx, strerror(err));
 }
 
+#ifndef CONFIG_BYTECODE_ONLY_RUNTIME
 static JSValue js_std_parseExtJSON(JSContext *ctx, JSValueConst this_val,
                                    int argc, JSValueConst *argv)
 {
@@ -971,6 +982,7 @@ static JSValue js_std_parseExtJSON(JSContext *ctx, JSValueConst this_val,
     JS_FreeCString(ctx, str);
     return obj;
 }
+#endif
 
 static JSValue js_new_std_file(JSContext *ctx, FILE *f,
                                BOOL close_in_finalizer,
@@ -1640,16 +1652,20 @@ static const JSCFunctionListEntry js_std_error_props[] = {
 static const JSCFunctionListEntry js_std_funcs[] = {
     JS_CFUNC_DEF("exit", 1, js_std_exit ),
     JS_CFUNC_DEF("gc", 0, js_std_gc ),
+#ifndef CONFIG_BYTECODE_ONLY_RUNTIME
     JS_CFUNC_DEF("evalScript", 1, js_evalScript ),
     JS_CFUNC_DEF("loadScript", 1, js_loadScript ),
+#endif
     JS_CFUNC_DEF("getenv", 1, js_std_getenv ),
     JS_CFUNC_DEF("setenv", 1, js_std_setenv ),
     JS_CFUNC_DEF("unsetenv", 1, js_std_unsetenv ),
     JS_CFUNC_DEF("getenviron", 1, js_std_getenviron ),
     JS_CFUNC_DEF("urlGet", 1, js_std_urlGet ),
     JS_CFUNC_DEF("loadFile", 1, js_std_loadFile ),
     JS_CFUNC_DEF("strerror", 1, js_std_strerror ),
+#ifndef CONFIG_BYTECODE_ONLY_RUNTIME
     JS_CFUNC_DEF("parseExtJSON", 1, js_std_parseExtJSON ),
+#endif
 
     /* FILE I/O */
     JS_CFUNC_DEF("open", 2, js_std_open ),
@@ -2268,7 +2284,7 @@ static void call_handler(JSContext *ctx, JSValueConst func)
     JS_FreeValue(ctx, ret);
 }
 
-#ifdef USE_WORKER
+#if defined(USE_WORKER) && !defined(CONFIG_BYTECODE_ONLY_RUNTIME)
 
 #ifdef _WIN32
 
@@ -3421,7 +3437,7 @@ static JSValue js_os_dup2(JSContext *ctx, JSValueConst this_val,
 
 #endif /* !_WIN32 */
 
-#ifdef USE_WORKER
+#if defined(USE_WORKER) && !defined(CONFIG_BYTECODE_ONLY_RUNTIME)
 
 /* Worker */
 
@@ -3610,7 +3626,7 @@ static void *worker_func(void *opaque)
 
     JS_SetCanBlock(rt, TRUE);
 
-    js_std_add_helpers(ctx, -1, NULL);
+    js_std_add_helpers(ctx, -1, NULL, TRUE);
 
     val = JS_LoadModule(ctx, args->basename, args->filename);
     free(args->filename);
@@ -3870,7 +3886,7 @@ static const JSCFunctionListEntry js_worker_proto_funcs[] = {
 
 void js_std_set_worker_new_context_func(JSContext *(*func)(JSRuntime *rt))
 {
-#ifdef USE_WORKER
+#if defined(USE_WORKER) && !defined(CONFIG_BYTECODE_ONLY_RUNTIME)
     js_worker_new_context_func = func;
 #endif
 }
@@ -3976,7 +3992,7 @@ static int js_os_init(JSContext *ctx, JSModuleDef *m)
 {
     os_poll_func = js_os_poll;
 
-#ifdef USE_WORKER
+#if defined(USE_WORKER) && !defined(CONFIG_BYTECODE_ONLY_RUNTIME)
     {
         JSRuntime *rt = JS_GetRuntime(ctx);
         JSThreadState *ts = JS_GetRuntimeOpaque(rt);
@@ -4015,7 +4031,7 @@ JSModuleDef *js_init_module_os(JSContext *ctx, const char *module_name)
     if (!m)
         return NULL;
     JS_AddModuleExportList(ctx, m, js_os_funcs, countof(js_os_funcs));
-#ifdef USE_WORKER
+#if defined(USE_WORKER) && !defined(CONFIG_BYTECODE_ONLY_RUNTIME)
     JS_AddModuleExport(ctx, m, "Worker");
 #endif
     return m;
@@ -4058,14 +4074,18 @@ static JSValue js_console_log(JSContext *ctx, JSValueConst this_val,
     return ret;
 }
 
-void js_std_add_helpers(JSContext *ctx, int argc, char **argv)
+void js_std_add_helpers(JSContext *ctx, int argc, char **argv, int use_module_loader)
 {
     JSValue global_obj, console, args, performance;
     int i;
 
     /* XXX: should these global definitions be enumerable? */
     global_obj = JS_GetGlobalObject(ctx);
 
+    if (use_module_loader) {
+        JS_SetModuleLoaderFunc2(JS_GetRuntime(ctx), NULL, js_module_loader, js_module_check_attributes, NULL);
+    }
+
     console = JS_NewObject(ctx);
     JS_SetPropertyStr(ctx, console, "log",
                       JS_NewCFunction(ctx, js_console_log, "log", 1));
@@ -4087,8 +4107,10 @@ void js_std_add_helpers(JSContext *ctx, int argc, char **argv)
 
     JS_SetPropertyStr(ctx, global_obj, "print",
                       JS_NewCFunction(ctx, js_print, "print", 1));
+#ifndef CONFIG_BYTECODE_ONLY_RUNTIME
     JS_SetPropertyStr(ctx, global_obj, "__loadScript",
                       JS_NewCFunction(ctx, js_loadScript, "__loadScript", 1));
+#endif
 
     JS_FreeValue(ctx, global_obj);
 }
@@ -4112,7 +4134,7 @@ void js_std_init_handlers(JSRuntime *rt)
 
     JS_SetRuntimeOpaque(rt, ts);
 
-#ifdef USE_WORKER
+#if defined(USE_WORKER) && !defined(CONFIG_BYTECODE_ONLY_RUNTIME)
     /* set the SharedArrayBuffer memory handlers */
     {
         JSSharedArrayBufferFunctions sf;
@@ -4152,7 +4174,7 @@ void js_std_free_handlers(JSRuntime *rt)
         free(rp);
     }
 
-#ifdef USE_WORKER
+#if defined(USE_WORKER) && !defined(CONFIG_BYTECODE_ONLY_RUNTIME)
     js_free_message_pipe(ts->recv_pipe);
     js_free_message_pipe(ts->send_pipe);
 

quickjs-libc.h

@@ -35,7 +35,7 @@ extern "C" {
 
 JSModuleDef *js_init_module_std(JSContext *ctx, const char *module_name);
 JSModuleDef *js_init_module_os(JSContext *ctx, const char *module_name);
-void js_std_add_helpers(JSContext *ctx, int argc, char **argv);
+void js_std_add_helpers(JSContext *ctx, int argc, char **argv, int use_module_loader);
 void js_std_loop(JSContext *ctx);
 JSValue js_std_await(JSContext *ctx, JSValue obj);
 void js_std_init_handlers(JSRuntime *rt);