Merge pull request #411 from toshok/tuple-colocate3

colocate string and tuple data

src/capi/descrobject.cpp

@@ -43,7 +43,7 @@ Box* BoxedMethodDescriptor::__call__(BoxedMethodDescriptor* self, Box* obj, Boxe
 
     Box* rtn;
     if (call_flags == METH_NOARGS) {
-        assert(varargs->elts.size() == 0);
+        assert(varargs->size() == 0);
         assert(kwargs->d.size() == 0);
         rtn = (Box*)self->method->ml_meth(obj, NULL);
     } else if (call_flags == METH_VARARGS) {
@@ -53,7 +53,7 @@ Box* BoxedMethodDescriptor::__call__(BoxedMethodDescriptor* self, Box* obj, Boxe
         rtn = (Box*)((PyCFunctionWithKeywords)self->method->ml_meth)(obj, varargs, kwargs);
     } else if (call_flags == METH_O) {
         assert(kwargs->d.size() == 0);
-        assert(varargs->elts.size() == 1);
+        assert(varargs->size() == 1);
         rtn = (Box*)self->method->ml_meth(obj, varargs->elts[0]);
     } else {
         RELEASE_ASSERT(0, "0x%x", call_flags);

src/capi/float.cpp

@@ -29,17 +29,17 @@ static float_format_type double_format, float_format;
 static float_format_type detected_double_format, detected_float_format;
 
 static PyObject* float_getformat(PyTypeObject* v, PyObject* arg) noexcept {
-    char* s;
     float_format_type r;
 
+    BoxedString* str = static_cast<BoxedString*>(arg);
+
     if (!PyString_Check(arg)) {
         PyErr_Format(PyExc_TypeError, "__getformat__() argument must be string, not %.500s", Py_TYPE(arg)->tp_name);
         return NULL;
     }
-    s = PyString_AS_STRING(arg);
-    if (strcmp(s, "double") == 0) {
+    if (str->s == "double") {
         r = double_format;
-    } else if (strcmp(s, "float") == 0) {
+    } else if (str->s == "float") {
         r = float_format;
     } else {
         PyErr_SetString(PyExc_ValueError, "__getformat__() argument 1 must be "
@@ -49,11 +49,11 @@ static PyObject* float_getformat(PyTypeObject* v, PyObject* arg) noexcept {
 
     switch (r) {
         case unknown_format:
-            return PyString_FromString("unknown");
+            return static_cast<PyObject*>(boxStrConstant("unknown"));
         case ieee_little_endian_format:
-            return PyString_FromString("IEEE, little-endian");
+            return static_cast<PyObject*>(boxStrConstant("IEEE, little-endian"));
         case ieee_big_endian_format:
-            return PyString_FromString("IEEE, big-endian");
+            return static_cast<PyObject*>(boxStrConstant("IEEE, big-endian"));
         default:
             Py_FatalError("insane float_format or double_format");
             return NULL;

src/capi/object.cpp

@@ -421,9 +421,9 @@ extern "C" PyObject* PyObject_SelfIter(PyObject* obj) noexcept {
 extern "C" int PyObject_GenericSetAttr(PyObject* obj, PyObject* name, PyObject* value) noexcept {
     try {
         if (value == NULL)
-            delattrGeneric(obj, static_cast<BoxedString*>(name)->s, NULL);
+            delattrGeneric(obj, std::string(static_cast<BoxedString*>(name)->s), NULL);
         else
-            setattrGeneric(obj, static_cast<BoxedString*>(name)->s.c_str(), value, NULL);
+            setattrGeneric(obj, std::string(static_cast<BoxedString*>(name)->s), value, NULL);
     } catch (ExcInfo e) {
         setCAPIException(e);
         return -1;

src/capi/typeobject.cpp

@@ -1577,7 +1577,7 @@ bool update_slot(BoxedClass* type, const std::string& attr) noexcept {
     }
     if (ptrs[0] == NULL)
         return false; /* Not an attribute that affects any slots */
-    int r = update_subclasses(type, new BoxedString(attr), update_slots_callback, (void*)ptrs);
+    int r = update_subclasses(type, boxString(attr), update_slots_callback, (void*)ptrs);
 
     // TODO this is supposed to be a CAPI function!
     if (r)
@@ -1634,13 +1634,13 @@ static PyObject* tp_new_wrapper(PyTypeObject* self, BoxedTuple* args, Box* kwds)
 
     RELEASE_ASSERT(args->cls == tuple_cls, "");
     RELEASE_ASSERT(kwds->cls == dict_cls, "");
-    RELEASE_ASSERT(args->elts.size() >= 1, "");
+    RELEASE_ASSERT(args->size() >= 1, "");
 
     BoxedClass* subtype = static_cast<BoxedClass*>(args->elts[0]);
     RELEASE_ASSERT(isSubclass(subtype->cls, type_cls), "");
     RELEASE_ASSERT(isSubclass(subtype, self), "");
 
-    BoxedTuple* new_args = new BoxedTuple(BoxedTuple::GCVector(args->elts.begin() + 1, args->elts.end()));
+    BoxedTuple* new_args = BoxedTuple::create(args->size() - 1, &args->elts[1]);
 
     return self->tp_new(subtype, new_args, kwds);
 }
@@ -2553,13 +2553,13 @@ static void remove_subclass(PyTypeObject* base, PyTypeObject* type) noexcept {
 void commonClassSetup(BoxedClass* cls) {
     if (cls->tp_bases == NULL) {
         if (cls->tp_base)
-            cls->tp_bases = new BoxedTuple({ cls->tp_base });
+            cls->tp_bases = BoxedTuple::create({ cls->tp_base });
         else
-            cls->tp_bases = new BoxedTuple({});
+            cls->tp_bases = BoxedTuple::create({});
     }
 
     /* Link into each base class's list of subclasses */
-    for (PyObject* b : static_cast<BoxedTuple*>(cls->tp_bases)->elts) {
+    for (PyObject* b : *static_cast<BoxedTuple*>(cls->tp_bases)) {
         if (PyType_Check(b) && add_subclass((PyTypeObject*)b, cls) < 0)
             throwCAPIException();
     }
@@ -2573,7 +2573,8 @@ void commonClassSetup(BoxedClass* cls) {
 
     assert(cls->tp_mro);
     assert(cls->tp_mro->cls == tuple_cls);
-    for (auto b : static_cast<BoxedTuple*>(cls->tp_mro)->elts) {
+
+    for (auto b : *static_cast<BoxedTuple*>(cls->tp_mro)) {
         if (b == cls)
             continue;
         if (PyType_Check(b))

src/capi/types.h

@@ -68,11 +68,11 @@ public:
             rtn = (Box*)((PyCFunctionWithKeywords)self->func)(self->passthrough, varargs, kwargs);
         } else if (self->ml_flags == METH_NOARGS) {
             assert(kwargs->d.size() == 0);
-            assert(varargs->elts.size() == 0);
+            assert(varargs->size() == 0);
             rtn = (Box*)self->func(self->passthrough, NULL);
         } else if (self->ml_flags == METH_O) {
             assert(kwargs->d.size() == 0);
-            assert(varargs->elts.size() == 1);
+            assert(varargs->size() == 1);
             rtn = (Box*)self->func(self->passthrough, varargs->elts[0]);
         } else {
             RELEASE_ASSERT(0, "0x%x", self->ml_flags);

src/codegen/ast_interpreter.cpp

@@ -580,7 +580,7 @@ Value ASTInterpreter::visit_langPrimitive(AST_LangPrimitive* node) {
         Box* type = last_exception.type;
         Box* value = last_exception.value ? last_exception.value : None;
         Box* traceback = last_exception.traceback ? last_exception.traceback : None;
-        v = new BoxedTuple({ type, value, traceback });
+        v = BoxedTuple::create({ type, value, traceback });
         last_exception = ExcInfo(NULL, NULL, NULL);
     } else if (node->opcode == AST_LangPrimitive::CHECK_EXC_MATCH) {
         assert(node->args.size() == 2);
@@ -747,11 +747,11 @@ Value ASTInterpreter::visit_makeClass(AST_MakeClass* mkclass) {
     ScopeInfo* scope_info = source_info->scoping->getScopeInfoForNode(node);
     assert(scope_info);
 
-    BoxedTuple::GCVector bases;
-    for (AST_expr* b : node->bases)
-        bases.push_back(visit_expr(b).o);
-
-    BoxedTuple* basesTuple = new BoxedTuple(std::move(bases));
+    BoxedTuple* basesTuple = BoxedTuple::create(node->bases.size());
+    int base_idx = 0;
+    for (AST_expr* b : node->bases) {
+        basesTuple->elts[base_idx++] = visit_expr(b).o;
+    }
 
     std::vector<Box*, StlCompatAllocator<Box*>> decorators;
     for (AST_expr* d : node->decorator_list)
@@ -1132,10 +1132,11 @@ Value ASTInterpreter::visit_list(AST_List* node) {
 }
 
 Value ASTInterpreter::visit_tuple(AST_Tuple* node) {
-    BoxedTuple::GCVector elts;
+    BoxedTuple* rtn = BoxedTuple::create(node->elts.size());
+    int rtn_idx = 0;
     for (AST_expr* e : node->elts)
-        elts.push_back(visit_expr(e).o);
-    return new BoxedTuple(std::move(elts));
+        rtn->elts[rtn_idx++] = visit_expr(e).o;
+    return rtn;
 }
 
 Value ASTInterpreter::visit_attribute(AST_Attribute* node) {
@@ -1194,7 +1195,7 @@ Box* astInterpretFrom(CompiledFunction* cf, AST_expr* after_expr, AST_stmt* encl
 
     for (const auto& p : frame_state.locals->d) {
         assert(p.first->cls == str_cls);
-        std::string name = static_cast<BoxedString*>(p.first)->s;
+        auto& name = static_cast<BoxedString*>(p.first)->s;
         if (name == PASSED_GENERATOR_NAME) {
             interpreter.setGenerator(p.second);
         } else if (name == PASSED_CLOSURE_NAME) {

src/codegen/compvars.cpp

@@ -2203,20 +2203,20 @@ public:
     Box* deserializeFromFrame(const FrameVals& vals) override {
         assert(vals.size() == numFrameArgs());
 
-        BoxedTuple::GCVector elts;
+        BoxedTuple* rtn = BoxedTuple::create(elt_types.size());
+        int rtn_idx = 0;
         int cur_idx = 0;
         for (auto e : elt_types) {
             int num_args = e->numFrameArgs();
             // TODO: inefficient to make these copies
             FrameVals sub_vals(vals.begin() + cur_idx, vals.begin() + cur_idx + num_args);
 
-            elts.push_back(e->deserializeFromFrame(sub_vals));
+            rtn->elts[rtn_idx++] = e->deserializeFromFrame(sub_vals);
 
             cur_idx += num_args;
         }
         assert(cur_idx == vals.size());
-
-        return new BoxedTuple(std::move(elts));
+        return rtn;
     }
 
     int numFrameArgs() override {

src/codegen/irgen/hooks.cpp

@@ -364,7 +364,7 @@ Box* eval(Box* boxedCode) {
     // TODO should have a cleaner interface that can parse the Expression directly
     // TODO this memory leaks
     RELEASE_ASSERT(boxedCode->cls == str_cls, "");
-    const char* code = static_cast<BoxedString*>(boxedCode)->s.c_str();
+    const char* code = static_cast<BoxedString*>(boxedCode)->s.data();
     AST_Module* parsedModule = parse_string(code);
     RELEASE_ASSERT(parsedModule->body[0]->type == AST_TYPE::Expr, "");
     AST_Expression* parsedExpr = new AST_Expression(std::move(parsedModule->interned_strings));
@@ -385,7 +385,7 @@ Box* exec(Box* boxedCode) {
 
     // TODO same issues as in `eval`
     RELEASE_ASSERT(boxedCode->cls == str_cls, "");
-    const char* code = static_cast<BoxedString*>(boxedCode)->s.c_str();
+    const char* code = static_cast<BoxedString*>(boxedCode)->s.data();
     AST_Module* parsedModule = parse_string(code);
     AST_Suite* parsedSuite = new AST_Suite(std::move(parsedModule->interned_strings));
     parsedSuite->body = parsedModule->body;

src/codegen/pypa-parser.cpp

@@ -850,12 +850,12 @@ pypa::String pypaUnicodeEscapeDecoder(pypa::String s, bool raw_prefix, bool& err
         checkAndThrowCAPIException();
         BoxedString* str_utf8 = (BoxedString*)PyUnicode_AsUTF8String(unicode);
         checkAndThrowCAPIException();
-        return str_utf8->s;
+        return std::string(str_utf8->s);
     } catch (ExcInfo e) {
         error = true;
         BoxedString* error_message = str(e.value);
         if (error_message && error_message->cls == str_cls)
-            return error_message->s;
+            return std::string(error_message->s);
         return "Encountered an unknown error inside pypaUnicodeEscapeDecoder";
     }
 }

src/jit.cpp

@@ -166,7 +166,7 @@ static int main(int argc, char** argv) {
     if (!Py_NoSiteFlag) {
         try {
             std::string module_name = "site";
-            importModuleLevel(&module_name, None, None, 0);
+            importModuleLevel(module_name, None, None, 0);
         } catch (ExcInfo e) {
             e.printExcAndTraceback();
             return 1;

src/runtime/builtin_modules/builtins.cpp

@@ -148,7 +148,7 @@ extern "C" Box* min(Box* arg0, BoxedTuple* args) {
 
     Box* minElement;
     Box* container;
-    if (args->elts.size() == 0) {
+    if (args->size() == 0) {
         minElement = nullptr;
         container = arg0;
     } else {
@@ -178,7 +178,7 @@ extern "C" Box* max(Box* arg0, BoxedTuple* args) {
 
     Box* maxElement;
     Box* container;
-    if (args->elts.size() == 0) {
+    if (args->size() == 0) {
         maxElement = nullptr;
         container = arg0;
     } else {
@@ -250,7 +250,8 @@ extern "C" Box* chr(Box* arg) {
         raiseExcHelper(ValueError, "chr() arg not in range(256)");
     }
 
-    return boxString(std::string(1, (char)n));
+    char c = (char)n;
+    return boxStringRef(llvm::StringRef(&c, 1));
 }
 
 extern "C" Box* unichr(Box* arg) {
@@ -382,7 +383,8 @@ Box* bltinImport(Box* name, Box* globals, Box* locals, Box** args) {
         raiseExcHelper(TypeError, "an integer is required");
     }
 
-    return importModuleLevel(&static_cast<BoxedString*>(name)->s, globals, fromlist, ((BoxedInt*)level)->n);
+    std::string _name = static_cast<BoxedString*>(name)->s;
+    return importModuleLevel(_name, globals, fromlist, ((BoxedInt*)level)->n);
 }
 
 Box* delattrFunc(Box* obj, Box* _str) {
@@ -391,7 +393,7 @@ Box* delattrFunc(Box* obj, Box* _str) {
     if (_str->cls != str_cls)
         raiseExcHelper(TypeError, "attribute name must be string, not '%s'", getTypeName(_str));
     BoxedString* str = static_cast<BoxedString*>(_str);
-    delattr(obj, str->s.c_str());
+    delattr(obj, str->s.data());
     return None;
 }
 
@@ -406,7 +408,7 @@ Box* getattrFunc(Box* obj, Box* _str, Box* default_value) {
 
     Box* rtn = NULL;
     try {
-        rtn = getattr(obj, str->s.c_str());
+        rtn = getattr(obj, str->s.data());
     } catch (ExcInfo e) {
         if (!e.matches(AttributeError))
             throw e;
@@ -416,7 +418,7 @@ Box* getattrFunc(Box* obj, Box* _str, Box* default_value) {
         if (default_value)
             return default_value;
         else
-            raiseExcHelper(AttributeError, "'%s' object has no attribute '%s'", getTypeName(obj), str->s.c_str());
+            raiseExcHelper(AttributeError, "'%s' object has no attribute '%s'", getTypeName(obj), str->s.data());
     }
 
     return rtn;
@@ -430,7 +432,7 @@ Box* setattrFunc(Box* obj, Box* _str, Box* value) {
     }
 
     BoxedString* str = static_cast<BoxedString*>(_str);
-    setattr(obj, str->s.c_str(), value);
+    setattr(obj, str->s.data(), value);
     return None;
 }
 
@@ -465,7 +467,7 @@ Box* map2(Box* f, Box* container) {
 
 Box* map(Box* f, BoxedTuple* args) {
     assert(args->cls == tuple_cls);
-    auto num_iterable = args->elts.size();
+    auto num_iterable = args->size();
     if (num_iterable < 1)
         raiseExcHelper(TypeError, "map() requires at least two args");
 
@@ -475,7 +477,8 @@ Box* map(Box* f, BoxedTuple* args) {
 
     std::vector<BoxIterator> args_it;
     std::vector<BoxIterator> args_end;
-    for (auto& e : args->elts) {
+
+    for (auto e : *args) {
         auto range = e->pyElements();
         args_it.emplace_back(range.begin());
         args_end.emplace_back(range.end());
@@ -559,8 +562,7 @@ Box* zip2(Box* container1, Box* container2) {
     BoxIterator it2 = range2.begin();
 
     for (; it1 != range1.end() && it2 != range2.end(); ++it1, ++it2) {
-        BoxedTuple::GCVector elts{ *it1, *it2 };
-        listAppendInternal(rtn, new BoxedTuple(std::move(elts)));
+        listAppendInternal(rtn, BoxedTuple::create({ *it1, *it2 }));
     }
     return rtn;
 }
@@ -583,7 +585,7 @@ public:
         RELEASE_ASSERT(isSubclass(self->cls, BaseException), "");
         BoxedException* exc = static_cast<BoxedException*>(self);
 
-        return new BoxedTuple({ self->cls, EmptyTuple, makeAttrWrapper(self) });
+        return BoxedTuple::create({ self->cls, EmptyTuple, makeAttrWrapper(self) });
     }
 };
 
@@ -598,7 +600,7 @@ Box* exceptionNew(BoxedClass* cls, BoxedTuple* args) {
     BoxedException* rtn = new (cls) BoxedException();
 
     // TODO: this should be a MemberDescriptor and set during init
-    if (args->elts.size() == 1)
+    if (args->size() == 1)
         rtn->giveAttr("message", args->elts[0]);
     else
         rtn->giveAttr("message", boxStrConstant(""));
@@ -623,7 +625,7 @@ Box* exceptionRepr(Box* b) {
     assert(message->cls == str_cls);
 
     BoxedString* message_s = static_cast<BoxedString*>(message);
-    return boxString(std::string(getTypeName(b)) + "(" + message_s->s + ",)");
+    return boxStringTwine(llvm::Twine(getTypeName(b)) + "(" + message_s->s + ",)");
 }
 
 static BoxedClass* makeBuiltinException(BoxedClass* base, const char* name, int size = 0) {
@@ -667,7 +669,7 @@ extern "C" PyObject* PyErr_NewException(char* name, PyObject* _base, PyObject* d
         }
         checkAndThrowCAPIException();
 
-        Box* cls = runtimeCall(type_cls, ArgPassSpec(3), boxedName, new BoxedTuple({ base }), dict, NULL, NULL);
+        Box* cls = runtimeCall(type_cls, ArgPassSpec(3), boxedName, BoxedTuple::create({ base }), dict, NULL, NULL);
         return cls;
     } catch (ExcInfo e) {
         abort();
@@ -706,7 +708,7 @@ public:
         BoxedEnumerate* self = static_cast<BoxedEnumerate*>(_self);
         Box* val = *self->iterator;
         ++self->iterator;
-        return new BoxedTuple({ boxInt(self->idx++), val });
+        return BoxedTuple::create({ boxInt(self->idx++), val });
     }
 
     static Box* hasnext(Box* _self) {
@@ -766,14 +768,14 @@ Box* execfile(Box* _fn) {
 #endif
 
 #else
-    bool exists = llvm::sys::fs::exists(fn->s);
+    bool exists = llvm::sys::fs::exists(std::string(fn->s));
 #endif
 
     if (!exists)
-        raiseExcHelper(IOError, "No such file or directory: '%s'", fn->s.c_str());
+        raiseExcHelper(IOError, "No such file or directory: '%s'", fn->s.data());
 
     // Run directly inside the current module:
-    AST_Module* ast = caching_parse_file(fn->s.c_str());
+    AST_Module* ast = caching_parse_file(fn->s.data());
     compileAndRunModule(ast, getCurrentModule());
 
     return None;
@@ -785,7 +787,7 @@ Box* print(BoxedTuple* args, BoxedDict* kwargs) {
 
     Box* dest, *end;
 
-    auto it = kwargs->d.find(new BoxedString("file"));
+    auto it = kwargs->d.find(boxStrConstant("file"));
     if (it != kwargs->d.end()) {
         dest = it->second;
         kwargs->d.erase(it);
@@ -793,23 +795,23 @@ Box* print(BoxedTuple* args, BoxedDict* kwargs) {
         dest = getSysStdout();
     }
 
-    it = kwargs->d.find(new BoxedString("end"));
+    it = kwargs->d.find(boxStrConstant("end"));
     if (it != kwargs->d.end()) {
         end = it->second;
         kwargs->d.erase(it);
     } else {
-        end = new BoxedString("\n");
+        end = boxStrConstant("\n");
     }
 
     RELEASE_ASSERT(kwargs->d.size() == 0, "print() got unexpected keyword arguments");
 
     static const std::string write_str("write");
 
-    Box* space_box = new BoxedString(" ");
+    Box* space_box = boxStrConstant(" ");
 
     // TODO softspace handling?
     bool first = true;
-    for (auto e : args->elts) {
+    for (auto e : *args) {
         BoxedString* s = str(e);
 
         if (!first) {

src/runtime/builtin_modules/pyston.cpp

@@ -43,7 +43,7 @@ static Box* setOption(Box* option, Box* value) {
     else CHECK(REOPT_THRESHOLD_BASELINE);
     else CHECK(OSR_THRESHOLD_BASELINE);
     else CHECK(SPECULATION_THRESHOLD);
-    else raiseExcHelper(ValueError, "unknown option name '%s", option_string->s.c_str());
+    else raiseExcHelper(ValueError, "unknown option name '%s", option_string->s.data());
 
     return None;
 }

src/runtime/builtin_modules/sys.cpp

@@ -43,7 +43,7 @@ Box* sysExcInfo() {
     assert(exc->type);
     assert(exc->value);
     assert(exc->traceback);
-    return new BoxedTuple({ exc->type, exc->value, exc->traceback });
+    return BoxedTuple::create({ exc->type, exc->value, exc->traceback });
 }
 
 Box* sysExcClear() {
@@ -184,7 +184,7 @@ void prependToSysPath(const std::string& path) {
     BoxedList* sys_path = getSysPath();
     static std::string attr = "insert";
     callattr(sys_path, &attr, CallattrFlags({.cls_only = false, .null_on_nonexistent = false }), ArgPassSpec(2),
-             boxInt(0), new BoxedString(path), NULL, NULL, NULL);
+             boxInt(0), boxString(path), NULL, NULL, NULL);
 }
 
 static BoxedClass* sys_flags_cls;
@@ -258,7 +258,7 @@ void setupSys() {
 
     sys_module->giveAttr("warnoptions", new BoxedList());
     sys_module->giveAttr("py3kwarning", False);
-    sys_module->giveAttr("byteorder", new BoxedString(isLittleEndian() ? "little" : "big"));
+    sys_module->giveAttr("byteorder", boxStrConstant(isLittleEndian() ? "little" : "big"));
 
     sys_module->giveAttr("platform", boxStrConstant("unknown")); // seems like a reasonable, if poor, default
 
@@ -293,14 +293,14 @@ void setupSys() {
     sys_module->giveAttr("hexversion", boxInt(PY_VERSION_HEX));
     // TODO: this should be a "sys.version_info" object, not just a tuple (ie can access fields by name)
     sys_module->giveAttr("version_info",
-                         new BoxedTuple({ boxInt(PYTHON_VERSION_MAJOR), boxInt(PYTHON_VERSION_MINOR),
-                                          boxInt(PYTHON_VERSION_MICRO), boxStrConstant("beta"), boxInt(0) }));
+                         BoxedTuple::create({ boxInt(PYTHON_VERSION_MAJOR), boxInt(PYTHON_VERSION_MINOR),
+                                              boxInt(PYTHON_VERSION_MICRO), boxStrConstant("beta"), boxInt(0) }));
 
     sys_module->giveAttr("maxint", boxInt(PYSTON_INT_MAX));
     sys_module->giveAttr("maxsize", boxInt(PY_SSIZE_T_MAX));
 
     sys_flags_cls = new BoxedHeapClass(object_cls, BoxedSysFlags::gcHandler, 0, 0, sizeof(BoxedSysFlags), false,
-                                       new BoxedString("flags"));
+                                       static_cast<BoxedString*>(boxString("flags")));
     sys_flags_cls->giveAttr("__new__",
                             new BoxedFunction(boxRTFunction((void*)BoxedSysFlags::__new__, UNKNOWN, 1, 0, true, true)));
 #define ADD(name)                                                                                                      \
@@ -311,7 +311,7 @@ void setupSys() {
     ADD(no_user_site);
 #undef ADD
 
-    sys_flags_cls->tp_mro = new BoxedTuple({ sys_flags_cls, object_cls });
+    sys_flags_cls->tp_mro = BoxedTuple::create({ sys_flags_cls, object_cls });
     sys_flags_cls->freeze();
 
     sys_module->giveAttr("flags", new BoxedSysFlags());
@@ -326,7 +326,8 @@ void setupSysEnd() {
     std::sort<decltype(builtin_module_names)::iterator, PyLt>(builtin_module_names.begin(), builtin_module_names.end(),
                                                               PyLt());
 
-    sys_module->giveAttr("builtin_module_names", new BoxedTuple(std::move(builtin_module_names)));
+    sys_module->giveAttr("builtin_module_names",
+                         BoxedTuple::create(builtin_module_names.size(), &builtin_module_names[0]));
     sys_flags_cls->finishInitialization();
 }
 }

src/runtime/capi.cpp

@@ -250,7 +250,7 @@ extern "C" PyObject* PyObject_GetAttr(PyObject* o, PyObject* attr_name) noexcept
     }
 
     try {
-        return getattr(o, static_cast<BoxedString*>(attr_name)->s.c_str());
+        return getattr(o, static_cast<BoxedString*>(attr_name)->data());
     } catch (ExcInfo e) {
         setCAPIException(e);
         return NULL;
@@ -258,7 +258,7 @@ extern "C" PyObject* PyObject_GetAttr(PyObject* o, PyObject* attr_name) noexcept
 }
 
 extern "C" PyObject* PyObject_GenericGetAttr(PyObject* o, PyObject* name) noexcept {
-    Box* r = getattrInternalGeneric(o, static_cast<BoxedString*>(name)->s.c_str(), NULL, false, false, NULL, NULL);
+    Box* r = getattrInternalGeneric(o, static_cast<BoxedString*>(name)->data(), NULL, false, false, NULL, NULL);
     if (!r)
         PyErr_Format(PyExc_AttributeError, "'%.50s' object has no attribute '%.400s'", o->cls->tp_name,
                      PyString_AS_STRING(name));
@@ -879,7 +879,8 @@ extern "C" PyObject* PyImport_Import(PyObject* module_name) noexcept {
     RELEASE_ASSERT(module_name->cls == str_cls, "");
 
     try {
-        return importModuleLevel(&static_cast<BoxedString*>(module_name)->s, None, None, -1);
+        std::string _module_name = static_cast<BoxedString*>(module_name)->s;
+        return importModuleLevel(_module_name, None, None, -1);
     } catch (ExcInfo e) {
         Py_FatalError("unimplemented");
     }

src/runtime/classobj.cpp

@@ -33,7 +33,7 @@ static Box* classLookup(BoxedClassobj* cls, const std::string& attr) {
     if (r)
         return r;
 
-    for (auto b : cls->bases->elts) {
+    for (auto b : *cls->bases) {
         RELEASE_ASSERT(b->cls == classobj_cls, "");
         Box* r = classLookup(static_cast<BoxedClassobj*>(b), attr);
         if (r)
@@ -88,7 +88,7 @@ Box* classobjNew(Box* _cls, Box* _name, Box* _bases, Box** _args) {
         raiseExcHelper(TypeError, "PyClass_New: bases must be a tuple");
     BoxedTuple* bases = static_cast<BoxedTuple*>(_bases);
 
-    for (auto base : bases->elts) {
+    for (auto base : *bases) {
         if (!PyClass_Check(base) && PyCallable_Check(base->cls)) {
             Box* r = PyObject_CallFunctionObjArgs(base->cls, name, bases, dict, NULL);
             if (!r)
@@ -104,7 +104,7 @@ Box* classobjNew(Box* _cls, Box* _name, Box* _bases, Box** _args) {
 
     for (auto& p : dict->d) {
         RELEASE_ASSERT(p.first->cls == str_cls, "");
-        made->setattr(static_cast<BoxedString*>(p.first)->s, p.second, NULL);
+        made->setattr(std::string(static_cast<BoxedString*>(p.first)->s), p.second, NULL);
     }
 
     // Note: make sure to do this after assigning the attrs, since it will overwrite any defined __name__
@@ -134,7 +134,7 @@ Box* classobjCall(Box* _cls, Box* _args, Box* _kwargs) {
         if (init_rtn != None)
             raiseExcHelper(TypeError, "__init__() should return None");
     } else {
-        if (args->elts.size() || kwargs->d.size())
+        if (args->size() || kwargs->d.size())
             raiseExcHelper(TypeError, "this constructor takes no arguments");
     }
     return made;
@@ -162,9 +162,9 @@ static Box* classobjGetattribute(Box* _cls, Box* _attr) {
         }
     }
 
-    Box* r = classLookup(cls, attr->s);
+    Box* r = classLookup(cls, std::string(attr->s));
     if (!r)
-        raiseExcHelper(AttributeError, "class %s has no attribute '%s'", cls->name->s.c_str(), attr->s.c_str());
+        raiseExcHelper(AttributeError, "class %s has no attribute '%s'", cls->name->data(), attr->data());
 
     r = processDescriptor(r, None, cls);
     return r;
@@ -242,7 +242,7 @@ Box* classobjStr(Box* _obj) {
     Box* _mod = cls->getattr("__module__");
     RELEASE_ASSERT(_mod, "");
     RELEASE_ASSERT(_mod->cls == str_cls, "");
-    return boxString(static_cast<BoxedString*>(_mod)->s + "." + cls->name->s);
+    return boxStringTwine(llvm::Twine(static_cast<BoxedString*>(_mod)->s) + "." + cls->name->s);
 }
 
 static Box* _instanceGetattribute(Box* _inst, Box* _attr, bool raise_on_missing) {
@@ -282,8 +282,7 @@ static Box* _instanceGetattribute(Box* _inst, Box* _attr, bool raise_on_missing)
     if (!raise_on_missing)
         return NULL;
 
-    raiseExcHelper(AttributeError, "%s instance has no attribute '%s'", inst->inst_cls->name->s.c_str(),
-                   attr->s.c_str());
+    raiseExcHelper(AttributeError, "%s instance has no attribute '%s'", inst->inst_cls->name->data(), attr->data());
 }
 
 Box* instanceGetattribute(Box* _inst, Box* _attr) {
@@ -361,7 +360,7 @@ Box* instanceRepr(Box* _inst) {
         assert(class_str->cls == str_cls);
 
         char buf[80];
-        snprintf(buf, 80, "<%s instance at %p>", static_cast<BoxedString*>(class_str)->s.c_str(), inst);
+        snprintf(buf, 80, "<%s instance at %p>", static_cast<BoxedString*>(class_str)->data(), inst);
         return boxStrConstant(buf);
     }
 }
@@ -495,7 +494,7 @@ Box* instanceCall(Box* _inst, Box* _args, Box* _kwargs) {
 
     Box* call_func = _instanceGetattribute(inst, boxStrConstant("__call__"), false);
     if (!call_func)
-        raiseExcHelper(AttributeError, "%s instance has no __call__ method", inst->inst_cls->name->s.c_str());
+        raiseExcHelper(AttributeError, "%s instance has no __call__ method", inst->inst_cls->name->data());
 
     return runtimeCall(call_func, ArgPassSpec(0, 0, true, true), _args, _kwargs, NULL, NULL, NULL);
 }

src/runtime/code.cpp

@@ -55,12 +55,12 @@ public:
 
         BoxedTuple::GCVector elts;
         for (auto sr : param_names.args)
-            elts.push_back(new BoxedString(sr));
+            elts.push_back(boxString(sr));
         if (param_names.vararg.size())
-            elts.push_back(new BoxedString(param_names.vararg));
+            elts.push_back(boxString(param_names.vararg));
         if (param_names.kwarg.size())
-            elts.push_back(new BoxedString(param_names.kwarg));
-        return new BoxedTuple(std::move(elts));
+            elts.push_back(boxString(param_names.kwarg));
+        return BoxedTuple::create(elts.size(), &elts[0]);
     }
 
     static Box* flags(Box* b, void*) {

src/runtime/dict.cpp

@@ -44,7 +44,7 @@ Box* dictRepr(BoxedDict* self) {
         chars.insert(chars.end(), v->s.begin(), v->s.end());
     }
     chars.push_back('}');
-    return boxString(std::string(chars.begin(), chars.end()));
+    return boxString(llvm::StringRef(&chars[0], chars.size()));
 }
 
 Box* dictClear(BoxedDict* self) {
@@ -69,10 +69,7 @@ Box* dictItems(BoxedDict* self) {
 
     rtn->ensure(self->d.size());
     for (const auto& p : self->d) {
-        BoxedTuple::GCVector elts;
-        elts.push_back(p.first);
-        elts.push_back(p.second);
-        BoxedTuple* t = new BoxedTuple(std::move(elts));
+        BoxedTuple* t = BoxedTuple::create({ p.first, p.second });
         listAppendInternal(rtn, t);
     }
 
@@ -378,7 +375,7 @@ Box* dictPopitem(BoxedDict* self) {
     Box* value = it->second;
     self->d.erase(it);
 
-    auto rtn = new BoxedTuple({ key, value });
+    auto rtn = BoxedTuple::create({ key, value });
     return rtn;
 }
 
@@ -506,9 +503,9 @@ void dictMergeFromSeq2(BoxedDict* self, Box* other) {
             self->d[list->elts->elts[0]] = list->elts->elts[1];
         } else if (element->cls == tuple_cls) {
             BoxedTuple* tuple = static_cast<BoxedTuple*>(element);
-            if (tuple->elts.size() != 2)
+            if (tuple->size() != 2)
                 raiseExcHelper(ValueError, "dictionary update sequence element #%d has length %d; 2 is required", idx,
-                               tuple->elts.size());
+                               tuple->size());
 
             self->d[tuple->elts[0]] = tuple->elts[1];
         } else
@@ -541,8 +538,8 @@ Box* dictUpdate(BoxedDict* self, BoxedTuple* args, BoxedDict* kwargs) {
     assert(kwargs);
     assert(kwargs->cls == dict_cls);
 
-    RELEASE_ASSERT(args->elts.size() <= 1, ""); // should throw a TypeError
-    if (args->elts.size()) {
+    RELEASE_ASSERT(args->size() <= 1, ""); // should throw a TypeError
+    if (args->size()) {
         Box* arg = args->elts[0];
         if (getattrInternal(arg, "keys", NULL)) {
             dictMerge(self, arg);
@@ -558,7 +555,7 @@ Box* dictUpdate(BoxedDict* self, BoxedTuple* args, BoxedDict* kwargs) {
 }
 
 extern "C" Box* dictInit(BoxedDict* self, BoxedTuple* args, BoxedDict* kwargs) {
-    int args_sz = args->elts.size();
+    int args_sz = args->size();
     int kwargs_sz = kwargs->d.size();
 
     // CPython accepts a single positional and keyword arguments, in any combination

src/runtime/file.cpp

@@ -195,7 +195,7 @@ BoxedFile::BoxedFile(FILE* f, std::string fname, const char* fmode, int (*close)
       f_bufptr(0),
       f_setbuf(0),
       unlocked_count(0) {
-    Box* r = fill_file_fields(this, f, new BoxedString(fname), fmode, close);
+    Box* r = fill_file_fields(this, f, boxString(fname), fmode, close);
     checkAndThrowCAPIException();
     assert(r == this);
 }
@@ -823,17 +823,17 @@ Box* fileNew(BoxedClass* cls, Box* s, Box* m) {
         raiseExcHelper(TypeError, "");
     }
 
-    const std::string& fn = static_cast<BoxedString*>(s)->s;
-    const std::string& mode = static_cast<BoxedString*>(m)->s;
+    auto fn = static_cast<BoxedString*>(s);
+    auto mode = static_cast<BoxedString*>(m);
 
-    FILE* f = fopen(fn.c_str(), mode.c_str());
+    FILE* f = fopen(fn->data(), mode->data());
     if (!f) {
-        PyErr_SetFromErrnoWithFilename(IOError, fn.c_str());
+        PyErr_SetFromErrnoWithFilename(IOError, fn->data());
         throwCAPIException();
         abort(); // unreachable;
     }
 
-    return new BoxedFile(f, fn, PyString_AsString(m));
+    return new BoxedFile(f, fn->s, PyString_AsString(m));
 }
 
 static PyObject* file_readlines(BoxedFile* f, PyObject* args) noexcept {

src/runtime/import.cpp

@@ -166,7 +166,7 @@ SearchResult findModule(const std::string& name, const std::string& full_name, B
         auto path_pass = path_list ? path_list : None;
         Box* loader
             = callattr(finder, &find_module_str, CallattrFlags({.cls_only = false, .null_on_nonexistent = false }),
-                       ArgPassSpec(2), new BoxedString(full_name), path_pass, NULL, NULL, NULL);
+                       ArgPassSpec(2), boxString(full_name), path_pass, NULL, NULL, NULL);
 
         if (loader != None)
             return SearchResult(loader);
@@ -195,7 +195,7 @@ SearchResult findModule(const std::string& name, const std::string& full_name, B
         BoxedString* p = static_cast<BoxedString*>(_p);
 
         joined_path.clear();
-        llvm::sys::path::append(joined_path, p->s, name);
+        llvm::sys::path::append(joined_path, std::string(p->s), name);
         std::string dn(joined_path.str());
 
         llvm::sys::path::append(joined_path, "__init__.py");
@@ -209,7 +209,7 @@ SearchResult findModule(const std::string& name, const std::string& full_name, B
             auto path_pass = path_list ? path_list : None;
             Box* loader = callattr(importer, &find_module_str,
                                    CallattrFlags({.cls_only = false, .null_on_nonexistent = false }), ArgPassSpec(2),
-                                   new BoxedString(full_name), path_pass, NULL, NULL, NULL);
+                                   boxString(full_name), path_pass, NULL, NULL, NULL);
             if (loader != None)
                 return SearchResult(loader);
         }
@@ -218,7 +218,7 @@ SearchResult findModule(const std::string& name, const std::string& full_name, B
             return SearchResult(std::move(dn), SearchResult::PKG_DIRECTORY);
 
         joined_path.clear();
-        llvm::sys::path::append(joined_path, p->s, name + ".py");
+        llvm::sys::path::append(joined_path, std::string(p->s), name + ".py");
         fn = joined_path.str();
 
         if (pathExists(fn))
@@ -250,7 +250,7 @@ static Box* getParent(Box* globals, int level, std::string& buf) {
         if (pkgname->cls != str_cls) {
             raiseExcHelper(ValueError, "__package__ set to non-string");
         }
-        size_t len = pkgname->s.size();
+        size_t len = pkgname->size();
         if (len == 0) {
             if (level > 0) {
                 raiseExcHelper(ValueError, "Attempted relative import in non-package");
@@ -270,7 +270,7 @@ static Box* getParent(Box* globals, int level, std::string& buf) {
         Box* modpath = getattrInternal(globals, path_str, NULL);
         if (modpath != NULL) {
             /* __path__ is set, so modname is already the package name */
-            if (modname->s.size() > PATH_MAX) {
+            if (modname->size() > PATH_MAX) {
                 raiseExcHelper(ValueError, "Module name too long");
             }
             buf += modname->s;
@@ -357,7 +357,7 @@ static Box* importSub(const std::string& name, const std::string& full_name, Box
             const static std::string load_module_str("load_module");
             module = callattr(sr.loader, &load_module_str,
                               CallattrFlags({.cls_only = false, .null_on_nonexistent = false }), ArgPassSpec(1),
-                              new BoxedString(full_name), NULL, NULL, NULL, NULL);
+                              boxString(full_name), NULL, NULL, NULL, NULL);
         } else
             RELEASE_ASSERT(0, "%d", sr.type);
 
@@ -441,9 +441,8 @@ static bool loadNext(Box* mod, Box* altmod, std::string& name, std::string& buf,
 }
 
 static void ensureFromlist(Box* module, Box* fromlist, std::string& buf, bool recursive);
-Box* importModuleLevel(std::string* name, Box* globals, Box* from_imports, int level) {
+Box* importModuleLevel(const std::string& name, Box* globals, Box* from_imports, int level) {
     bool return_first = from_imports == None;
-    assert(name);
 
     static StatCounter slowpath_import("slowpath_import");
     slowpath_import.log();
@@ -453,7 +452,9 @@ Box* importModuleLevel(std::string* name, Box* globals, Box* from_imports, int l
     if (!parent)
         return NULL;
 
-    std::string _name = *name;
+    // make a copy of the string we pass in, since loadNext while
+    // modify the string as we load the modules along the path.
+    std::string _name = name;
 
     Box* head;
     bool again = loadNext(parent, level < 0 ? NULL : parent, _name, buf, &head);
@@ -505,7 +506,7 @@ extern "C" PyObject* PyImport_ImportModuleLevel(const char* name, PyObject* glob
                                                 PyObject* fromlist, int level) noexcept {
     try {
         std::string module_name = name;
-        return importModuleLevel(&module_name, globals, fromlist ? fromlist : None, level);
+        return importModuleLevel(module_name, globals, fromlist ? fromlist : None, level);
     } catch (ExcInfo e) {
         setCAPIException(e);
         return NULL;
@@ -539,7 +540,7 @@ static void ensureFromlist(Box* module, Box* fromlist, std::string& buf, bool re
             continue;
 
         // Just want to import it and add it to the modules list for now:
-        importSub(s->s, buf + '.' + s->s, module);
+        importSub(s->s, (llvm::Twine(buf) + "." + s->s).str(), module);
     }
 }
 
@@ -586,7 +587,7 @@ extern "C" PyObject* PyImport_ExecCodeModuleEx(char* name, PyObject* co, char* p
         if (module == NULL)
             return NULL;
 
-        AST_Module* ast = parse_string(code->s.c_str());
+        AST_Module* ast = parse_string(code->data());
         compileAndRunModule(ast, module);
         module->setattr("__file__", boxString(pathname), NULL);
         return module;
@@ -611,7 +612,7 @@ Box* nullImporterInit(Box* self, Box* _path) {
     if (path->s.empty())
         raiseExcHelper(ImportError, "empty pathname");
 
-    if (isdir(path->s.c_str()))
+    if (isdir(path->data()))
         raiseExcHelper(ImportError, "existing directory");
 
     return None;
@@ -623,7 +624,7 @@ Box* nullImporterFindModule(Box* self, Box* fullname, Box* path) {
 
 extern "C" Box* import(int level, Box* from_imports, const std::string* module_name) {
     std::string _module_name(*module_name);
-    return importModuleLevel(&_module_name, getCurrentModule(), from_imports, level);
+    return importModuleLevel(_module_name, getCurrentModule(), from_imports, level);
 }
 
 Box* impFindModule(Box* _name, BoxedList* path) {
@@ -635,19 +636,19 @@ Box* impFindModule(Box* _name, BoxedList* path) {
 
     SearchResult sr = findModule(name->s, name->s, path_list);
     if (sr.type == SearchResult::SEARCH_ERROR)
-        raiseExcHelper(ImportError, "%s", name->s.c_str());
+        raiseExcHelper(ImportError, "%s", name->data());
 
     if (sr.type == SearchResult::PY_SOURCE) {
         Box* path = boxString(sr.path);
         Box* mode = boxStrConstant("r");
         Box* f = runtimeCall(file_cls, ArgPassSpec(2), path, mode, NULL, NULL, NULL);
-        return new BoxedTuple({ f, path, new BoxedTuple({ boxStrConstant(".py"), mode, boxInt(sr.type) }) });
+        return BoxedTuple::create({ f, path, BoxedTuple::create({ boxStrConstant(".py"), mode, boxInt(sr.type) }) });
     }
 
     if (sr.type == SearchResult::PKG_DIRECTORY) {
         Box* path = boxString(sr.path);
-        Box* mode = boxStrConstant("");
-        return new BoxedTuple({ None, path, new BoxedTuple({ mode, mode, boxInt(sr.type) }) });
+        Box* mode = EmptyString;
+        return BoxedTuple::create({ None, path, BoxedTuple::create({ mode, mode, boxInt(sr.type) }) });
     }
 
     Py_FatalError("unimplemented");
@@ -665,7 +666,7 @@ Box* impLoadModule(Box* _name, Box* _file, Box* _pathname, Box** args) {
     BoxedString* pathname = (BoxedString*)_pathname;
     BoxedTuple* description = (BoxedTuple*)_description;
 
-    RELEASE_ASSERT(description->elts.size() == 3, "");
+    RELEASE_ASSERT(description->size() == 3, "");
     BoxedString* suffix = (BoxedString*)description->elts[0];
     BoxedString* mode = (BoxedString*)description->elts[1];
     BoxedInt* type = (BoxedInt*)description->elts[2];
@@ -678,7 +679,7 @@ Box* impLoadModule(Box* _name, Box* _file, Box* _pathname, Box** args) {
     RELEASE_ASSERT(mode->s.empty(), "");
 
     if (type->n == SearchResult::PKG_DIRECTORY) {
-        return createAndRunModule(name->s, pathname->s + "/__init__.py", pathname->s);
+        return createAndRunModule(name->s, (llvm::Twine(pathname->s) + "/__init__.py").str(), pathname->s);
     }
 
     Py_FatalError("unimplemented");
@@ -688,7 +689,7 @@ Box* impGetSuffixes() {
     BoxedList* list = new BoxedList;
     // For now only add *.py
     listAppendInternal(
-        list, new BoxedTuple({ new BoxedString(".py"), new BoxedString("U"), boxInt(SearchResult::PY_SOURCE) }));
+        list, BoxedTuple::create({ boxStrConstant(".py"), boxStrConstant("U"), boxInt(SearchResult::PY_SOURCE) }));
     return list;
 }
 

src/runtime/import.h

@@ -20,7 +20,7 @@
 namespace pyston {
 
 extern "C" Box* import(int level, Box* from_imports, const std::string* module_name);
-extern Box* importModuleLevel(std::string* module_name, Box* globals, Box* from_imports, int level);
+extern Box* importModuleLevel(const std::string& module_name, Box* globals, Box* from_imports, int level);
 }
 
 #endif

src/runtime/inline/boxing.cpp

@@ -14,6 +14,8 @@
 
 #include "runtime/inline/boxing.h"
 
+#include "llvm/ADT/SmallString.h"
+
 #include "runtime/int.h"
 #include "runtime/objmodel.h"
 #include "runtime/types.h"
@@ -28,25 +30,43 @@ extern "C" Box* createList() {
     return new BoxedList();
 }
 
+BoxedString* _boxStrConstant(const char* chars) {
+    size_t len = strlen(chars);
+    return new (len) BoxedString(chars, len);
+}
 extern "C" BoxedString* boxStrConstant(const char* chars) {
-    return new BoxedString(chars);
+    return _boxStrConstant(chars);
+}
+
+BoxedString* _boxStrConstantSize(const char* chars, size_t n) {
+    return new (n) BoxedString(chars, n);
 }
 
 extern "C" BoxedString* boxStrConstantSize(const char* chars, size_t n) {
-    return new BoxedString(chars, n);
+    return _boxStrConstantSize(chars, n);
 }
 
 extern "C" Box* boxStringPtr(const std::string* s) {
-    return new BoxedString(*s);
+    return new (s->size()) BoxedString(s->c_str(), s->size());
+}
+
+Box* boxStringRef(llvm::StringRef s) {
+    return new (s.size()) BoxedString(s);
 }
 
 Box* boxString(const std::string& s) {
-    return new BoxedString(s);
+    return new (s.size()) BoxedString(s.c_str(), s.size());
 }
 Box* boxString(std::string&& s) {
-    return new BoxedString(std::move(s));
+    return new (s.size()) BoxedString(s.c_str(), s.size());
+}
+
+Box* boxStringTwine(const llvm::Twine& t) {
+    llvm::SmallString<256> Vec;
+    return boxString(t.toStringRef(Vec));
 }
 
+
 extern "C" double unboxFloat(Box* b) {
     ASSERT(b->cls == float_cls, "%s", getTypeName(b));
     BoxedFloat* f = (BoxedFloat*)b;

src/runtime/inline/dict.cpp

@@ -66,8 +66,7 @@ Box* dictIterNext(Box* s) {
     } else if (self->type == BoxedDictIterator::ValueIterator) {
         rtn = self->it->second;
     } else if (self->type == BoxedDictIterator::ItemIterator) {
-        BoxedTuple::GCVector elts{ self->it->first, self->it->second };
-        rtn = new BoxedTuple(std::move(elts));
+        rtn = BoxedTuple::create({ self->it->first, self->it->second });
     }
     ++self->it;
     return rtn;

src/runtime/inline/tuple.cpp

@@ -40,14 +40,14 @@ i1 tupleiterHasnextUnboxed(Box* s) {
     assert(s->cls == tuple_iterator_cls);
     BoxedTupleIterator* self = static_cast<BoxedTupleIterator*>(s);
 
-    return self->pos < self->t->elts.size();
+    return self->pos < self->t->size();
 }
 
 Box* tupleiterNext(Box* s) {
     assert(s->cls == tuple_iterator_cls);
     BoxedTupleIterator* self = static_cast<BoxedTupleIterator*>(s);
 
-    if (!(self->pos >= 0 && self->pos < self->t->elts.size())) {
+    if (!(self->pos >= 0 && self->pos < self->t->size())) {
         raiseExcHelper(StopIteration, "");
     }
 

src/runtime/int.cpp

@@ -873,7 +873,7 @@ extern "C" BoxedString* intRepr(BoxedInt* v) {
 
     char buf[80];
     int len = snprintf(buf, 80, "%ld", v->n);
-    return new BoxedString(std::string(buf, len));
+    return static_cast<BoxedString*>(boxString(llvm::StringRef(buf, len)));
 }
 
 extern "C" Box* intHash(BoxedInt* self) {
@@ -898,7 +898,7 @@ extern "C" Box* intHex(BoxedInt* self) {
         len = snprintf(buf, sizeof(buf), "-0x%lx", std::abs(self->n));
     else
         len = snprintf(buf, sizeof(buf), "0x%lx", self->n);
-    return new BoxedString(std::string(buf, len));
+    return boxStringRef(llvm::StringRef(buf, len));
 }
 
 extern "C" Box* intOct(BoxedInt* self) {
@@ -913,7 +913,7 @@ extern "C" Box* intOct(BoxedInt* self) {
         len = snprintf(buf, sizeof(buf), "-%#lo", std::abs(self->n));
     else
         len = snprintf(buf, sizeof(buf), "%#lo", self->n);
-    return new BoxedString(std::string(buf, len));
+    return boxString(llvm::StringRef(buf, len));
 }
 
 extern "C" Box* intTrunc(BoxedInt* self) {
@@ -942,8 +942,8 @@ static Box* _intNew(Box* val, Box* base) {
 
         BoxedString* s = static_cast<BoxedString*>(val);
 
-        RELEASE_ASSERT(s->s.size() == strlen(s->s.c_str()), "");
-        Box* r = PyInt_FromString(s->s.c_str(), NULL, base_n);
+        RELEASE_ASSERT(s->size() == strlen(s->data()), "");
+        Box* r = PyInt_FromString(s->data(), NULL, base_n);
         if (!r)
             throwCAPIException();
         return r;

src/runtime/iterators.cpp

@@ -80,11 +80,11 @@ private:
     static bool hasnext(BoxedList* o, uint64_t i) { return i < o->size; }
     static Box* getValue(BoxedList* o, uint64_t i) { return o->elts->elts[i]; }
 
-    static bool hasnext(BoxedTuple* o, uint64_t i) { return i < o->elts.size(); }
+    static bool hasnext(BoxedTuple* o, uint64_t i) { return i < o->size(); }
     static Box* getValue(BoxedTuple* o, uint64_t i) { return o->elts[i]; }
 
-    static bool hasnext(BoxedString* o, uint64_t i) { return i < o->s.size(); }
-    static Box* getValue(BoxedString* o, uint64_t i) { return new BoxedString(std::string(1, o->s[i])); }
+    static bool hasnext(BoxedString* o, uint64_t i) { return i < o->size(); }
+    static Box* getValue(BoxedString* o, uint64_t i) { return boxString(llvm::StringRef(o->data() + i, 1)); }
 
 public:
     BoxIteratorIndex(T* obj) : obj(obj), index(0) {

src/runtime/list.cpp

@@ -16,7 +16,8 @@
 
 #include <algorithm>
 #include <cstring>
-#include <sstream>
+
+#include "llvm/Support/raw_ostream.h"
 
 #include "capi/types.h"
 #include "core/ast.h"
@@ -55,14 +56,16 @@ extern "C" PyObject* PyList_AsTuple(PyObject* v) noexcept {
     }
 
     auto l = static_cast<BoxedList*>(v);
-    return new BoxedTuple(BoxedTuple::GCVector(l->elts->elts, l->elts->elts + l->size));
+    return BoxedTuple::create(l->size, &l->elts->elts[0]);
 }
 
 extern "C" Box* listRepr(BoxedList* self) {
     LOCK_REGION(self->lock.asRead());
 
     // TODO highly inefficient with all the string copying
-    std::ostringstream os;
+
+    std::string O("");
+    llvm::raw_string_ostream os(O);
     os << '[';
     for (int i = 0; i < self->size; i++) {
         if (i > 0)
@@ -75,7 +78,7 @@ extern "C" Box* listRepr(BoxedList* self) {
         os << s->s;
     }
     os << ']';
-    return new BoxedString(os.str());
+    return boxString(os.str());
 }
 
 extern "C" Box* listNonzero(BoxedList* self) {
@@ -282,7 +285,7 @@ extern "C" Box* listSetitemSlice(BoxedList* self, BoxedSlice* slice, Box* v) {
     if (!v) {
         v_size = 0;
         v_elts = NULL;
-    } else if (isSubclass(v->cls, list_cls)) {
+    } else if (v->cls == list_cls || isSubclass(v->cls, list_cls)) {
         BoxedList* lv = static_cast<BoxedList*>(v);
         v_size = lv->size;
         // If lv->size is 0, lv->elts->elts is garbage
@@ -290,9 +293,9 @@ extern "C" Box* listSetitemSlice(BoxedList* self, BoxedSlice* slice, Box* v) {
             v_elts = lv->elts->elts;
         else
             v_elts = NULL;
-    } else if (isSubclass(v->cls, tuple_cls)) {
+    } else if (v->cls == tuple_cls || isSubclass(v->cls, tuple_cls)) {
         BoxedTuple* tv = static_cast<BoxedTuple*>(v);
-        v_size = tv->elts.size();
+        v_size = tv->size();
         v_elts = &tv->elts[0];
     } else {
         RELEASE_ASSERT(0, "unsupported type for list slice assignment: '%s'", getTypeName(v));
@@ -517,7 +520,7 @@ void listSort(BoxedList* self, Box* cmp, Box* key, Box* reverse) {
                 // original object.
                 // TODO we could potentially make this faster by copying the CPython approach of
                 // creating special sortwrapper objects that compare only based on the key.
-                Box* new_obj = new BoxedTuple({ key_val, boxInt(i), *obj_loc });
+                Box* new_obj = BoxedTuple::create({ key_val, boxInt(i), *obj_loc });
 
                 *obj_loc = new_obj;
                 num_keys_added++;
@@ -629,7 +632,7 @@ Box* listIndex(BoxedList* self, Box* elt, BoxedInt* _start, Box** args) {
     }
 
     BoxedString* tostr = static_cast<BoxedString*>(repr(elt));
-    raiseExcHelper(ValueError, "%s is not in list", tostr->s.c_str());
+    raiseExcHelper(ValueError, "%s is not in list", tostr->data());
 }
 
 Box* listRemove(BoxedList* self, Box* elt) {

src/runtime/long.cpp

@@ -268,7 +268,7 @@ extern "C" PyAPI_FUNC(PyObject*) _PyLong_Format(PyObject* aa, int base, int addL
         os << "L";
 
     os.flush();
-    auto rtn = new BoxedString(std::move(str));
+    auto rtn = boxString(str);
     free(buf);
     return rtn;
 }
@@ -451,7 +451,7 @@ BoxedLong* _longNew(Box* val, Box* _base) {
             raiseExcHelper(TypeError, "long() arg2 must be >= 2 and <= 36");
         }
 
-        int r = mpz_init_set_str(rtn->n, s->s.c_str(), base);
+        int r = mpz_init_set_str(rtn->n, s->data(), base);
         RELEASE_ASSERT(r == 0, "");
     } else {
         if (isSubclass(val->cls, long_cls)) {
@@ -936,7 +936,7 @@ extern "C" Box* longDivmod(BoxedLong* lhs, Box* _rhs) {
         mpz_init(q->n);
         mpz_init(r->n);
         mpz_fdiv_qr(q->n, r->n, lhs->n, rhs->n);
-        return new BoxedTuple({ q, r });
+        return BoxedTuple::create({ q, r });
     } else if (isSubclass(_rhs->cls, int_cls)) {
         BoxedInt* rhs = static_cast<BoxedInt*>(_rhs);
 
@@ -948,7 +948,7 @@ extern "C" Box* longDivmod(BoxedLong* lhs, Box* _rhs) {
         mpz_init(q->n);
         mpz_init_set_si(r->n, rhs->n);
         mpz_fdiv_qr(q->n, r->n, lhs->n, r->n);
-        return new BoxedTuple({ q, r });
+        return BoxedTuple::create({ q, r });
     } else {
         return NotImplemented;
     }

src/runtime/objmodel.cpp

@@ -128,8 +128,9 @@ static Box* (*callattrInternal3)(Box*, const std::string*, LookupScope, CallRewr
 
 size_t PyHasher::operator()(Box* b) const {
     if (b->cls == str_cls) {
-        StringHash<std::string> H;
-        return H(static_cast<BoxedString*>(b)->s);
+        StringHash<char> H;
+        auto s = static_cast<BoxedString*>(b);
+        return H(s->data(), s->size());
     }
 
     BoxedInt* i = hash(b);
@@ -218,7 +219,7 @@ extern "C" bool isSubclass(BoxedClass* child, BoxedClass* parent) {
 extern "C" void assertFail(BoxedModule* inModule, Box* msg) {
     if (msg) {
         BoxedString* tostr = str(msg);
-        raiseExcHelper(AssertionError, "%s", tostr->s.c_str());
+        raiseExcHelper(AssertionError, "%s", tostr->data());
     } else {
         raiseExcHelper(AssertionError, "");
     }
@@ -272,7 +273,7 @@ static void _checkUnpackingLength(i64 expected, i64 given) {
 extern "C" Box** unpackIntoArray(Box* obj, int64_t expected_size) {
     if (obj->cls == tuple_cls) {
         BoxedTuple* t = static_cast<BoxedTuple*>(obj);
-        _checkUnpackingLength(expected_size, t->elts.size());
+        _checkUnpackingLength(expected_size, t->size());
         return &t->elts[0];
     }
 
@@ -413,7 +414,7 @@ BoxedHeapClass::BoxedHeapClass(BoxedClass* base, gcvisit_func gc_visit, int attr
     if (!ht_name)
         assert(str_cls == NULL);
     else
-        tp_name = ht_name->s.c_str();
+        tp_name = ht_name->data();
 
     memset(&as_number, 0, sizeof(as_number));
     memset(&as_mapping, 0, sizeof(as_mapping));
@@ -425,7 +426,7 @@ BoxedHeapClass* BoxedHeapClass::create(BoxedClass* metaclass, BoxedClass* base,
                                        int weaklist_offset, int instance_size, bool is_user_defined,
                                        const std::string& name) {
     return create(metaclass, base, gc_visit, attrs_offset, weaklist_offset, instance_size, is_user_defined,
-                  new BoxedString(name), NULL);
+                  static_cast<BoxedString*>(boxString(name)), NULL);
 }
 
 BoxedHeapClass* BoxedHeapClass::create(BoxedClass* metaclass, BoxedClass* base, gcvisit_func gc_visit, int attrs_offset,
@@ -458,7 +459,7 @@ std::string getFullNameOfClass(BoxedClass* cls) {
 
     BoxedString* module = static_cast<BoxedString*>(b);
 
-    return module->s + "." + cls->tp_name;
+    return (llvm::Twine(module->s) + "." + cls->tp_name).str();
 }
 
 std::string getFullTypeName(Box* o) {
@@ -782,7 +783,7 @@ Box* typeLookup(BoxedClass* cls, const std::string& attr, GetattrRewriteArgs* re
         // address.
         obj_saved->addAttrGuard(offsetof(BoxedClass, tp_mro), (intptr_t)mro);
 
-        for (auto base : mro->elts) {
+        for (auto base : *mro) {
             rewrite_args->out_success = false;
             if (base == cls) {
                 // Small optimization: don't have to load the class again since it was given to us in
@@ -801,7 +802,7 @@ Box* typeLookup(BoxedClass* cls, const std::string& attr, GetattrRewriteArgs* re
     } else {
         assert(cls->tp_mro);
         assert(cls->tp_mro->cls == tuple_cls);
-        for (auto b : static_cast<BoxedTuple*>(cls->tp_mro)->elts) {
+        for (auto b : *static_cast<BoxedTuple*>(cls->tp_mro)) {
             val = b->getattr(attr, NULL);
             if (val)
                 return val;
@@ -952,7 +953,7 @@ Box* descriptorClsSpecialCases(GetattrRewriteArgs* rewrite_args, BoxedClass* cls
 Box* boxChar(char c) {
     char d[1];
     d[0] = c;
-    return new BoxedString(std::string(d, 1));
+    return boxStringRef(llvm::StringRef(d, 1));
 }
 
 static Box* noneIfNull(Box* b) {
@@ -967,12 +968,12 @@ static Box* boxStringOrNone(const char* s) {
     if (s == NULL) {
         return None;
     } else {
-        return boxString(std::string(s));
+        return boxStrConstant(s);
     }
 }
 
 static Box* boxStringFromCharPtr(const char* s) {
-    return boxString(std::string(s));
+    return boxStrConstant(s);
 }
 
 Box* dataDescriptorInstanceSpecialCases(GetattrRewriteArgs* rewrite_args, const std::string& attr_name, Box* obj,
@@ -1100,7 +1101,7 @@ Box* dataDescriptorInstanceSpecialCases(GetattrRewriteArgs* rewrite_args, const
                 rewrite_args = NULL;
                 REWRITE_ABORTED("");
                 char* rtn = reinterpret_cast<char*>((char*)obj + member_desc->offset);
-                return boxString(std::string(rtn));
+                return boxStringRef(llvm::StringRef(rtn));
             }
 
             default:
@@ -1164,7 +1165,7 @@ Box* getattrInternalEx(Box* obj, const std::string& attr, GetattrRewriteArgs* re
     if (!cls_only) {
         BoxedClass* cls = obj->cls;
         if (obj->cls->tp_getattro && obj->cls->tp_getattro != PyObject_GenericGetAttr) {
-            Box* r = obj->cls->tp_getattro(obj, new BoxedString(attr));
+            Box* r = obj->cls->tp_getattro(obj, boxString(attr));
             if (!r)
                 throwCAPIException();
             return r;
@@ -2205,7 +2206,7 @@ extern "C" void dump(void* p) {
 
             if (cls->tp_mro && cls->tp_mro->cls == tuple_cls) {
                 bool first = true;
-                for (auto b : static_cast<BoxedTuple*>(cls->tp_mro)->elts) {
+                for (auto b : *static_cast<BoxedTuple*>(cls->tp_mro)) {
                     if (!first)
                         printf(" ->");
                     first = false;
@@ -2216,11 +2217,11 @@ extern "C" void dump(void* p) {
         }
 
         if (isSubclass(b->cls, str_cls)) {
-            printf("String value: %s\n", static_cast<BoxedString*>(b)->s.c_str());
+            printf("String value: %s\n", static_cast<BoxedString*>(b)->data());
         }
 
         if (isSubclass(b->cls, tuple_cls)) {
-            printf("%ld elements\n", static_cast<BoxedTuple*>(b)->elts.size());
+            printf("%ld elements\n", static_cast<BoxedTuple*>(b)->size());
         }
 
         if (isSubclass(b->cls, int_cls)) {
@@ -2760,7 +2761,7 @@ Box* callFunc(BoxedFunctionBase* func, CallRewriteArgs* rewrite_args, ArgPassSpe
                 rewrite_args->args->setAttr((varargs_idx - 3) * sizeof(Box*), emptyTupleConst);
         }
 
-        Box* ovarargs = new BoxedTuple(BoxedTuple::GCVector(unused_positional.begin(), unused_positional.end()));
+        Box* ovarargs = BoxedTuple::create(unused_positional.size(), &unused_positional[0]);
         getArg(varargs_idx, oarg1, oarg2, oarg3, oargs) = ovarargs;
     } else if (unused_positional.size()) {
         raiseExcHelper(TypeError, "%s() takes at most %d argument%s (%d given)", getFunctionName(f).c_str(),
@@ -2851,7 +2852,7 @@ Box* callFunc(BoxedFunctionBase* func, CallRewriteArgs* rewrite_args, ArgPassSpe
                 Box*& v = okwargs->d[p.first];
                 if (v) {
                     raiseExcHelper(TypeError, "%s() got multiple values for keyword argument '%s'",
-                                   getFunctionName(f).c_str(), s->s.c_str());
+                                   getFunctionName(f).c_str(), s->data());
                 }
                 v = p.second;
             }
@@ -3840,14 +3841,15 @@ Box* typeNew(Box* _cls, Box* arg1, Box* arg2, Box** _args) {
     RELEASE_ASSERT(arg1->cls == str_cls, "");
     BoxedString* name = static_cast<BoxedString*>(arg1);
 
-    if (bases->elts.size() == 0) {
-        bases = new BoxedTuple({ object_cls });
+    if (bases->size() == 0) {
+        bases = BoxedTuple::create({ object_cls });
     }
 
     // Ported from CPython:
-    int nbases = bases->elts.size();
+    int nbases = bases->size();
     BoxedClass* winner = metatype;
-    for (auto tmp : bases->elts) {
+
+    for (auto tmp : *bases) {
         auto tmptype = tmp->cls;
         if (tmptype == classobj_cls)
             continue;
@@ -3926,7 +3928,7 @@ Box* typeNew(Box* _cls, Box* arg1, Box* arg2, Box** _args) {
         made->tp_alloc = PyType_GenericAlloc;
 
     assert(!made->simple_destructor);
-    for (auto b : bases->elts) {
+    for (auto b : *bases) {
         if (!isSubclass(b->cls, type_cls))
             continue;
         BoxedClass* b_cls = static_cast<BoxedClass*>(b);
@@ -4228,7 +4230,7 @@ Box* typeCallInternal(BoxedFunctionBase* f, CallRewriteArgs* rewrite_args, ArgPa
 Box* typeCall(Box* obj, BoxedTuple* vararg, BoxedDict* kwargs) {
     assert(vararg->cls == tuple_cls);
 
-    int n = vararg->elts.size();
+    int n = vararg->size();
     int args_to_pass = n + 2; // 1 for obj, 1 for kwargs
 
     Box** args = NULL;
@@ -4371,7 +4373,7 @@ extern "C" Box* importStar(Box* _from_module, BoxedModule* to_module) {
             Box* attr_value = from_module->getattr(casted_attr_name->s);
 
             if (!attr_value)
-                raiseExcHelper(AttributeError, "'module' object has no attribute '%s'", casted_attr_name->s.c_str());
+                raiseExcHelper(AttributeError, "'module' object has no attribute '%s'", casted_attr_name->data());
 
             to_module->setattr(casted_attr_name->s, attr_value, NULL);
         }

src/runtime/set.cpp

@@ -14,7 +14,7 @@
 
 #include "runtime/set.h"
 
-#include <sstream>
+#include <llvm/Support/raw_ostream.h>
 
 #include "gc/collector.h"
 #include "runtime/objmodel.h"
@@ -97,7 +97,8 @@ Box* setNew(Box* _cls, Box* container) {
 }
 
 static Box* _setRepr(BoxedSet* self, const char* type_name) {
-    std::ostringstream os("");
+    std::string O("");
+    llvm::raw_string_ostream os(O);
 
     os << type_name << "([";
     bool first = true;
@@ -232,7 +233,7 @@ Box* setUpdate(BoxedSet* self, BoxedTuple* args) {
     assert(self->cls == set_cls);
     assert(args->cls == tuple_cls);
 
-    for (auto l : args->elts) {
+    for (auto l : *args) {
         if (l->cls == set_cls) {
             BoxedSet* s2 = static_cast<BoxedSet*>(l);
             self->s.insert(s2->s.begin(), s2->s.end());

src/runtime/super.cpp

@@ -51,7 +51,7 @@ public:
     }
 };
 
-static const std::string class_str("__class__");
+static const char* class_str = "__class__";
 
 Box* superGetattribute(Box* _s, Box* _attr) {
     RELEASE_ASSERT(_s->cls == super_cls, "");
@@ -101,7 +101,7 @@ Box* superGetattribute(Box* _s, Box* _attr) {
                 continue;
             res = PyDict_GetItem(dict, name);
 #endif
-            res = tmp->getattr(attr->s);
+            res = tmp->getattr(std::string(attr->s));
 
             if (res != NULL) {
 // Pyston change:
@@ -128,7 +128,7 @@ Box* superGetattribute(Box* _s, Box* _attr) {
         }
     }
 
-    Box* r = typeLookup(s->cls, attr->s, NULL);
+    Box* r = typeLookup(s->cls, std::string(attr->s), NULL);
     // TODO implement this
     RELEASE_ASSERT(r, "should call the equivalent of objectGetattr here");
     return processDescriptor(r, s, s->cls);
@@ -139,10 +139,11 @@ Box* superRepr(Box* _s) {
     BoxedSuper* s = static_cast<BoxedSuper*>(_s);
 
     if (s->obj_type) {
-        return boxString("<super: <class '" + std::string(s->type ? getNameOfClass(s->type) : "NULL") + "'>, <"
-                         + std::string(getNameOfClass(s->obj_type)) + " object>>");
+        return boxStringTwine(llvm::Twine("<super: <class '") + (s->type ? getNameOfClass(s->type) : "NULL") + "'>, <"
+                              + getNameOfClass(s->obj_type) + " object>>");
     } else {
-        return boxString("<super: <class '" + std::string(s->type ? getNameOfClass(s->type) : "NULL") + "'>, <NULL>>");
+        return boxStringTwine(llvm::Twine("<super: <class '") + (s->type ? getNameOfClass(s->type) : "NULL")
+                              + "'>, <NULL>>");
     }
 }
 

src/runtime/traceback.cpp

@@ -98,7 +98,7 @@ Box* BoxedTraceback::getLines(Box* b) {
         BoxedList* lines = new BoxedList();
         lines->ensure(tb->lines.size());
         for (auto line : tb->lines) {
-            auto l = new BoxedTuple({ boxString(line->file), boxString(line->func), boxInt(line->line) });
+            auto l = BoxedTuple::create({ boxString(line->file), boxString(line->func), boxInt(line->line) });
             listAppendInternal(lines, l);
         }
         tb->py_lines = lines;

src/runtime/tuple.cpp

@@ -15,7 +15,8 @@
 #include "runtime/tuple.h"
 
 #include <algorithm>
-#include <sstream>
+
+#include "llvm/Support/raw_ostream.h"
 
 #include "core/ast.h"
 #include "core/common.h"
@@ -29,13 +30,12 @@
 namespace pyston {
 
 extern "C" Box* createTuple(int64_t nelts, Box** elts) {
-    BoxedTuple::GCVector velts(elts, elts + nelts);
-    return new BoxedTuple(std::move(velts));
+    return BoxedTuple::create(nelts, elts);
 }
 
 Box* _tupleSlice(BoxedTuple* self, i64 start, i64 stop, i64 step, i64 length) {
 
-    i64 size = self->elts.size();
+    i64 size = self->size();
     assert(step != 0);
     if (step > 0) {
         assert(0 <= start);
@@ -45,15 +45,15 @@ Box* _tupleSlice(BoxedTuple* self, i64 start, i64 stop, i64 step, i64 length) {
         assert(-1 <= stop);
     }
 
-    // FIXME: No need to initialize with 0.
-    BoxedTuple::GCVector velts(length, 0);
+    // FIXME: No need to initialize with NULL, since we're going to fill it in
+    auto rtn = BoxedTuple::create(length);
     if (length > 0)
-        copySlice(&velts[0], &self->elts[0], start, step, length);
-    return new BoxedTuple(std::move(velts));
+        copySlice(&rtn->elts[0], &self->elts[0], start, step, length);
+    return rtn;
 }
 
 Box* tupleGetitemUnboxed(BoxedTuple* self, i64 n) {
-    i64 size = self->elts.size();
+    i64 size = self->size();
 
     if (n < 0)
         n = size + n;
@@ -89,7 +89,7 @@ Box* tupleGetitemSlice(BoxedTuple* self, BoxedSlice* slice) {
     assert(slice->cls == slice_cls);
 
     i64 start, stop, step, length;
-    parseSlice(slice, self->elts.size(), &start, &stop, &step, &length);
+    parseSlice(slice, self->size(), &start, &stop, &step, &length);
     return _tupleSlice(self, start, stop, step, length);
 }
 
@@ -97,7 +97,7 @@ extern "C" PyObject* PyTuple_GetSlice(PyObject* p, Py_ssize_t low, Py_ssize_t hi
     RELEASE_ASSERT(isSubclass(p->cls, tuple_cls), "");
     BoxedTuple* t = static_cast<BoxedTuple*>(p);
 
-    Py_ssize_t n = t->elts.size();
+    Py_ssize_t n = t->size();
     if (low < 0)
         low = 0;
     if (high > n)
@@ -108,7 +108,39 @@ extern "C" PyObject* PyTuple_GetSlice(PyObject* p, Py_ssize_t low, Py_ssize_t hi
     if (low == 0 && high == n)
         return p;
 
-    return new BoxedTuple(BoxedTuple::GCVector(&t->elts[low], &t->elts[high]));
+    return BoxedTuple::create(high - low, &t->elts[low]);
+}
+
+extern "C" int _PyTuple_Resize(PyObject** pv, Py_ssize_t newsize) noexcept {
+    // This is only allowed to be called when there is only one user of the tuple (ie a refcount of 1 in CPython)
+    assert(pv);
+    return BoxedTuple::Resize((BoxedTuple**)pv, newsize);
+}
+
+int BoxedTuple::Resize(BoxedTuple** pv, size_t newsize) noexcept {
+    assert(isSubclass((*pv)->cls, tuple_cls));
+
+    BoxedTuple* t = static_cast<BoxedTuple*>(*pv);
+
+    if (newsize == t->size())
+        return 0;
+
+    if (newsize < t->size()) {
+        // XXX resize the box (by reallocating) smaller if it makes sense
+        t->nelts = newsize;
+        return 0;
+    }
+
+    BoxedTuple* resized;
+
+    if (t->cls == tuple_cls)
+        resized = new (newsize) BoxedTuple(newsize); // we want an uninitialized tuple, but this will memset it with 0.
+    else
+        resized = new (t->cls, newsize) BoxedTuple(newsize); // we need an uninitialized string, but this will memset
+    memmove(resized->elts, t->elts, t->size());
+
+    *pv = resized;
+    return 0;
 }
 
 Box* tupleGetitem(BoxedTuple* self, Box* slice) {
@@ -128,10 +160,11 @@ Box* tupleAdd(BoxedTuple* self, Box* rhs) {
     }
 
     BoxedTuple* _rhs = static_cast<BoxedTuple*>(rhs);
-    BoxedTuple::GCVector velts;
-    velts.insert(velts.end(), self->elts.begin(), self->elts.end());
-    velts.insert(velts.end(), _rhs->elts.begin(), _rhs->elts.end());
-    return new BoxedTuple(std::move(velts));
+
+    BoxedTuple* rtn = BoxedTuple::create(self->size() + _rhs->size());
+    memmove(&rtn->elts[0], &self->elts[0], self->size() * sizeof(Box*));
+    memmove(&rtn->elts[self->size()], &_rhs->elts[0], _rhs->size() * sizeof(Box*));
+    return rtn;
 }
 
 Box* tupleMul(BoxedTuple* self, Box* rhs) {
@@ -140,7 +173,7 @@ Box* tupleMul(BoxedTuple* self, Box* rhs) {
     }
 
     int n = static_cast<BoxedInt*>(rhs)->n;
-    int s = self->elts.size();
+    int s = self->size();
 
     if (n < 0)
         n = 0;
@@ -148,33 +181,34 @@ Box* tupleMul(BoxedTuple* self, Box* rhs) {
     if (s == 0 || n == 1) {
         return self;
     } else {
-        BoxedTuple::GCVector velts(n * s);
-        auto iter = velts.begin();
+        BoxedTuple* rtn = BoxedTuple::create(n * s);
+        int rtn_i = 0;
         for (int i = 0; i < n; ++i) {
-            std::copy(self->elts.begin(), self->elts.end(), iter);
-            iter += s;
+            memmove(&rtn->elts[rtn_i], &self->elts[0], sizeof(Box*) * s);
+            rtn_i += s;
         }
-        return new BoxedTuple(std::move(velts));
+        return rtn;
     }
 }
 
 Box* tupleLen(BoxedTuple* t) {
     assert(isSubclass(t->cls, tuple_cls));
-    return boxInt(t->elts.size());
+    return boxInt(t->size());
 }
 
 extern "C" Py_ssize_t PyTuple_Size(PyObject* op) noexcept {
     RELEASE_ASSERT(PyTuple_Check(op), "");
-    return static_cast<BoxedTuple*>(op)->elts.size();
+    return static_cast<BoxedTuple*>(op)->size();
 }
 
 Box* tupleRepr(BoxedTuple* t) {
     assert(isSubclass(t->cls, tuple_cls));
 
-    std::ostringstream os("");
+    std::string O("");
+    llvm::raw_string_ostream os(O);
     os << "(";
 
-    int n = t->elts.size();
+    int n = t->size();
     for (int i = 0; i < n; i++) {
         if (i)
             os << ", ";
@@ -190,8 +224,8 @@ Box* tupleRepr(BoxedTuple* t) {
 }
 
 Box* _tupleCmp(BoxedTuple* lhs, BoxedTuple* rhs, AST_TYPE::AST_TYPE op_type) {
-    int lsz = lhs->elts.size();
-    int rsz = rhs->elts.size();
+    int lsz = lhs->size();
+    int rsz = rhs->size();
 
     bool is_order
         = (op_type == AST_TYPE::Lt || op_type == AST_TYPE::LtE || op_type == AST_TYPE::Gt || op_type == AST_TYPE::GtE);
@@ -274,11 +308,11 @@ Box* tupleNe(BoxedTuple* self, Box* rhs) {
 
 Box* tupleNonzero(BoxedTuple* self) {
     RELEASE_ASSERT(isSubclass(self->cls, tuple_cls), "");
-    return boxBool(self->elts.size() != 0);
+    return boxBool(self->size() != 0);
 }
 
 Box* tupleContains(BoxedTuple* self, Box* elt) {
-    int size = self->elts.size();
+    int size = self->size();
     for (int i = 0; i < size; i++) {
         Box* e = self->elts[i];
         Box* cmp = compareInternal(e, elt, AST_TYPE::Eq, NULL);
@@ -290,7 +324,7 @@ Box* tupleContains(BoxedTuple* self, Box* elt) {
 }
 
 Box* tupleIndex(BoxedTuple* self, Box* elt) {
-    int size = self->elts.size();
+    int size = self->size();
     for (int i = 0; i < size; i++) {
         Box* e = self->elts[i];
         Box* cmp = compareInternal(e, elt, AST_TYPE::Eq, NULL);
@@ -306,7 +340,7 @@ Box* tupleHash(BoxedTuple* self) {
     assert(isSubclass(self->cls, tuple_cls));
 
     int64_t rtn = 3527539;
-    for (Box* e : self->elts) {
+    for (auto e : *self) {
         BoxedInt* h = hash(e);
         assert(isSubclass(h->cls, int_cls));
         rtn ^= h->n + 0x9e3779b9 + (rtn << 6) + (rtn >> 2);
@@ -324,16 +358,14 @@ extern "C" Box* tupleNew(Box* _cls, BoxedTuple* args, BoxedDict* kwargs) {
         raiseExcHelper(TypeError, "tuple.__new__(%s): %s is not a subtype of tuple", getNameOfClass(cls),
                        getNameOfClass(cls));
 
-    int args_sz = args->elts.size();
+    int args_sz = args->size();
     int kwargs_sz = kwargs->d.size();
 
     if (args_sz + kwargs_sz > 1)
         raiseExcHelper(TypeError, "tuple() takes at most 1 argument (%d given)", args_sz + kwargs_sz);
 
-    BoxedTuple::GCVector velts;
-    Box* elements;
-
     if (args_sz || kwargs_sz) {
+        Box* elements;
         // if initializing from iterable argument, check common case positional args first
         if (args_sz) {
             elements = args->elts[0];
@@ -345,21 +377,24 @@ extern "C" Box* tupleNew(Box* _cls, BoxedTuple* args, BoxedDict* kwargs) {
             if (kw->s == "sequence")
                 elements = seq.second;
             else
-                raiseExcHelper(TypeError, "'%s' is an invalid keyword argument for this function", kw->s.c_str());
+                raiseExcHelper(TypeError, "'%s' is an invalid keyword argument for this function", kw->data());
         }
 
+        std::vector<Box*, StlCompatAllocator<Box*>> elts;
         for (auto e : elements->pyElements())
-            velts.push_back(e);
-    }
+            elts.push_back(e);
 
-    return new (cls) BoxedTuple(std::move(velts));
+        return BoxedTuple::create(elts.size(), &elts[0], cls);
+    } else {
+        return BoxedTuple::create(0, cls);
+    }
 }
 
 extern "C" int PyTuple_SetItem(PyObject* op, Py_ssize_t i, PyObject* newitem) noexcept {
     RELEASE_ASSERT(PyTuple_Check(op), "");
 
     BoxedTuple* t = static_cast<BoxedTuple*>(op);
-    RELEASE_ASSERT(i >= 0 && i < t->elts.size(), "");
+    RELEASE_ASSERT(i >= 0 && i < t->size(), "");
     t->elts[i] = newitem;
     return 0;
 }
@@ -386,7 +421,7 @@ extern "C" PyObject* PyTuple_Pack(Py_ssize_t n, ...) noexcept {
 extern "C" PyObject* PyTuple_New(Py_ssize_t size) noexcept {
     RELEASE_ASSERT(size >= 0, "");
 
-    return new BoxedTuple(BoxedTuple::GCVector(size, NULL));
+    return BoxedTuple::create(size);
 }
 
 

src/runtime/types.h

@@ -16,6 +16,7 @@
 #define PYSTON_RUNTIME_TYPES_H
 
 #include <llvm/ADT/StringMap.h>
+#include <llvm/ADT/Twine.h>
 #include <ucontext.h>
 
 #include "Python.h"
@@ -110,6 +111,9 @@ extern "C" Box* boxUnboundInstanceMethod(Box* func);
 extern "C" Box* boxStringPtr(const std::string* s);
 Box* boxString(const std::string& s);
 Box* boxString(std::string&& s);
+Box* boxStringRef(llvm::StringRef s);
+Box* boxStringTwine(const llvm::Twine& s);
+
 extern "C" BoxedString* boxStrConstant(const char* chars);
 extern "C" BoxedString* boxStrConstantSize(const char* chars, size_t n);
 extern "C" Box* decodeUTF8StringPtr(const std::string* s);
@@ -371,14 +375,34 @@ public:
 
 class BoxedString : public Box {
 public:
-    // const std::basic_string<char, std::char_traits<char>, StlCompatAllocator<char> > s;
-    std::string s;
+    llvm::StringRef s;
+
+    char* data() { return const_cast<char*>(s.data()); }
+    size_t size() { return s.size(); }
+
+    void* operator new(size_t size, size_t ssize) __attribute__((visibility("default"))) {
+        Box* rtn = static_cast<Box*>(gc_alloc(str_cls->tp_basicsize + ssize + 1, gc::GCKind::PYTHON));
+        rtn->cls = str_cls;
+        return rtn;
+    }
 
+    void* operator new(size_t size, BoxedClass* cls, size_t ssize) __attribute__((visibility("default"))) {
+        Box* rtn = static_cast<Box*>(cls->tp_alloc(cls, ssize + 1));
+        rtn->cls = cls;
+        return rtn;
+    }
+
+    // these should be private, but strNew needs them
     BoxedString(const char* s, size_t n) __attribute__((visibility("default")));
-    BoxedString(std::string&& s) __attribute__((visibility("default")));
-    BoxedString(const std::string& s) __attribute__((visibility("default")));
+    explicit BoxedString(size_t n, char c) __attribute__((visibility("default")));
+    explicit BoxedString(llvm::StringRef s) __attribute__((visibility("default")));
+    explicit BoxedString(llvm::StringRef lhs, llvm::StringRef rhs) __attribute__((visibility("default")));
+
+private:
+    // used only in ctors to give our llvm::StringRef the proper pointer
+    char* storage() { return (char*)this + cls->tp_basicsize; }
 
-    DEFAULT_CLASS_SIMPLE(str_cls);
+    void* operator new(size_t size) = delete;
 };
 
 class BoxedUnicode : public Box {
@@ -464,14 +488,65 @@ public:
 class BoxedTuple : public Box {
 public:
     typedef std::vector<Box*, StlCompatAllocator<Box*>> GCVector;
-    GCVector elts;
 
-    BoxedTuple(GCVector& elts) __attribute__((visibility("default"))) : elts(elts) {}
-    BoxedTuple(GCVector&& elts) __attribute__((visibility("default"))) : elts(std::move(elts)) {}
+    Box** elts;
+
+    void* operator new(size_t size, size_t nelts) __attribute__((visibility("default"))) {
+        Box* rtn = static_cast<Box*>(gc_alloc(_PyObject_VAR_SIZE(tuple_cls, nelts + 1), gc::GCKind::PYTHON));
+        rtn->cls = tuple_cls;
+        return rtn;
+    }
+
+    void* operator new(size_t size, BoxedClass* cls, size_t nelts) __attribute__((visibility("default"))) {
+        Box* rtn = static_cast<Box*>(cls->tp_alloc(cls, nelts));
+        rtn->cls = cls;
+        return rtn;
+    }
+
+    static BoxedTuple* create(int64_t size) { return new (size) BoxedTuple(size); }
+    static BoxedTuple* create(int64_t nelts, Box** elts) {
+        BoxedTuple* rtn = new (nelts) BoxedTuple(nelts);
+        memmove(&rtn->elts[0], elts, sizeof(Box*) * nelts);
+        return rtn;
+    }
+    static BoxedTuple* create(std::initializer_list<Box*> members) { return new (members.size()) BoxedTuple(members); }
+
+    static BoxedTuple* create(int64_t size, BoxedClass* cls) { return new (cls, size) BoxedTuple(size); }
+    static BoxedTuple* create(int64_t nelts, Box** elts, BoxedClass* cls) {
+        BoxedTuple* rtn = new (cls, nelts) BoxedTuple(nelts);
+        memmove(&rtn->elts[0], elts, sizeof(Box*) * nelts);
+        return rtn;
+    }
+    static BoxedTuple* create(std::initializer_list<Box*> members, BoxedClass* cls) {
+        return new (cls, members.size()) BoxedTuple(members);
+    }
+
+    static int Resize(BoxedTuple** pt, size_t newsize) noexcept;
+
+    Box** begin() const { return &elts[0]; }
+    Box** end() const { return &elts[nelts]; }
+
+    size_t size() const { return nelts; }
 
-    DEFAULT_CLASS_SIMPLE(tuple_cls);
+private:
+    size_t nelts;
+
+    BoxedTuple(size_t size) : elts(reinterpret_cast<Box**>((char*)this + this->cls->tp_basicsize)), nelts(size) {
+        memset(elts, 0, sizeof(Box*) * size);
+    }
+
+    BoxedTuple(std::initializer_list<Box*>& members)
+        : elts(reinterpret_cast<Box**>((char*)this + this->cls->tp_basicsize)), nelts(members.size()) {
+        // by the time we make it here elts[] is big enough to contain members
+        Box** p = &elts[0];
+        for (auto b : members) {
+            *p++ = b;
+        }
+    }
 };
+
 extern "C" BoxedTuple* EmptyTuple;
+extern "C" BoxedString* EmptyString;
 
 struct PyHasher {
     size_t operator()(Box*) const;
@@ -735,4 +810,5 @@ extern Box* dict_descr;
 
 Box* codeForFunction(BoxedFunction*);
 }
+
 #endif