fix: 原生盒 mark 边收集去对象预过滤——盒内字符串/BigInt 边并入 live 标记闭合 sweep 悬垂

This commit is contained in:
2026-09-12 18:10:03 +00:00
parent e368751f59
commit 473f73a1f6
9 changed files with 349 additions and 90 deletions
@@ -533,7 +533,8 @@ pub fn get_capability_ptr(obj: &JsObject) -> *mut DisposeCapability {
obj.native_data() as *mut DisposeCapability
}
/// 收集栈内作为对象引用的 value/method(GC 根边),供跨 epoch 遍历/重写时追踪。
/// 收集栈内全部 value/method 边(GC mark 边):对象/字符串/BigInt 均产出,
/// 消费侧按值类型分发到对象栈与存活集。
pub fn dispose_edges(obj: &JsObject) -> Vec<JsValue> {
if !obj.is_disposable_stack_obj() && !obj.is_async_disposable_stack_obj() {
return Vec::new();
@@ -544,14 +545,7 @@ pub fn dispose_edges(obj: &JsObject) -> Vec<JsValue> {
}
// SAFETY: native_data 持有 `alloc_capability` 写入的有效 Box 指针,本函数
// 只在对象存活期间被 GC/绑定层调用。
unsafe {
(*ptr)
.entries
.iter()
.flat_map(|entry| [entry.value, entry.method])
.filter(|value: &JsValue| value.is_object())
.collect()
}
unsafe { (*ptr).entries.iter().flat_map(|entry| [entry.value, entry.method]).collect() }
}
/// 克隆状态盒到新对象,用 `rewrite` 改写其中的对象引用
+3 -8
View File
@@ -68,7 +68,8 @@ fn alloc_map<H: VmHost>(vm: &mut H) -> *mut JsObject {
vm.alloc_object(obj)
}
/// 收集 Map 中作为对象引用的键和值(GC 根边),供跨 epoch 遍历/重写时追踪。
/// 收集 Map 内全部键值边(GC mark 边):对象/字符串/BigInt 均产出,
/// 消费侧按值类型分发到对象栈与存活集。
pub fn map_native_edges(obj: &JsObject) -> Vec<JsValue> {
if !obj.is_map() {
return Vec::new();
@@ -77,13 +78,7 @@ pub fn map_native_edges(obj: &JsObject) -> Vec<JsValue> {
if inner.is_null() {
return Vec::new();
}
unsafe {
(*inner)
.iter()
.flat_map(|(key, value)| [key.0, *value])
.filter(|value: &JsValue| value.is_object())
.collect()
}
unsafe { (*inner).iter().flat_map(|(key, value)| [key.0, *value]).collect() }
}
/// 克隆 Map 的 native 数据到新对象,用 `rewrite` 改写其中的对象引用
+3 -8
View File
@@ -110,7 +110,8 @@ fn alloc_set<H: VmHost>(vm: &mut H) -> *mut JsObject {
vm.alloc_object(obj)
}
/// 收集 Set 中作为对象引用的元素(GC 根边),供跨 epoch 遍历/重写时追踪。
/// 收集 Set 内全部元素边(GC mark 边):对象/字符串/BigInt 均产出,
/// 消费侧按值类型分发到对象栈与存活集。
pub fn set_native_edges(obj: &JsObject) -> Vec<JsValue> {
if !obj.is_set() {
return Vec::new();
@@ -119,13 +120,7 @@ pub fn set_native_edges(obj: &JsObject) -> Vec<JsValue> {
if inner.is_null() {
return Vec::new();
}
unsafe {
(*inner)
.iter()
.map(|key| key.0)
.filter(|value: &JsValue| value.is_object())
.collect()
}
unsafe { (*inner).iter().map(|key| key.0).collect() }
}
/// 克隆 Set 的 native 数据到新对象,用 `rewrite` 改写其中的对象引用。
+9 -17
View File
@@ -492,28 +492,20 @@ pub(crate) fn async_holds_suspended_for_in(obj: &JsObject) -> bool {
!state.suspended.for_in_iters.is_empty()
}
/// 异步状态内所有对象引用的扁平列表(GC mark 边)。
/// 异步状态内全部引用边的扁平列表(GC mark 边):对象/字符串/BigInt 均产出,
/// 消费侧按值类型分发到对象栈与存活集。
pub(crate) fn async_native_edges(obj: &JsObject) -> Vec<JsValue> {
let Some(state) = async_state_mut(obj) else {
return Vec::new();
};
let mut edges = Vec::new();
let push = |v: JsValue, edges: &mut Vec<JsValue>| {
if v.is_object() {
edges.push(v);
}
};
push(state.callee, &mut edges);
edges.extend(state.args.iter().copied().filter(|v| v.is_object()));
push(state.result, &mut edges);
push(state.promise, &mut edges);
push(state.resolve, &mut edges);
push(state.reject, &mut edges);
state.suspended.for_each_value(|v| {
if v.is_object() {
edges.push(v);
}
});
edges.push(state.callee);
edges.extend(state.args.iter().copied());
edges.push(state.result);
edges.push(state.promise);
edges.push(state.resolve);
edges.push(state.reject);
state.suspended.for_each_value(|v| edges.push(v));
edges
}
+12 -20
View File
@@ -878,35 +878,27 @@ pub(crate) fn async_generator_holds_suspended_for_in(obj: &JsObject) -> bool {
!state.suspended.for_in_iters.is_empty()
}
/// 异步生成器状态内所有对象引用的扁平列表(GC mark 边)。
/// 异步生成器状态内全部引用边的扁平列表(GC mark 边):对象/字符串/BigInt
/// 均产出,消费侧按值类型分发到对象栈与存活集。
pub(crate) fn async_generator_native_edges(obj: &JsObject) -> Vec<JsValue> {
let Some(state) = async_gen_state_mut(obj) else {
return Vec::new();
};
let mut edges = Vec::new();
let push = |v: JsValue, edges: &mut Vec<JsValue>| {
if v.is_object() {
edges.push(v);
}
};
push(state.callee, &mut edges);
edges.extend(state.args.iter().copied().filter(|v| v.is_object()));
push(state.result, &mut edges);
edges.push(state.callee);
edges.extend(state.args.iter().copied());
edges.push(state.result);
for req in state.queue.iter() {
push(req.promise, &mut edges);
push(req.resolve, &mut edges);
push(req.reject, &mut edges);
edges.push(req.promise);
edges.push(req.resolve);
edges.push(req.reject);
}
if let Some(req) = &state.current {
push(req.promise, &mut edges);
push(req.resolve, &mut edges);
push(req.reject, &mut edges);
edges.push(req.promise);
edges.push(req.resolve);
edges.push(req.reject);
}
state.suspended.for_each_value(|v| {
if v.is_object() {
edges.push(v);
}
});
state.suspended.for_each_value(|v| edges.push(v));
edges
}
+6 -14
View File
@@ -873,25 +873,17 @@ pub(crate) fn generator_holds_suspended_for_in(obj: &JsObject) -> bool {
!state.suspended.for_in_iters.is_empty()
}
/// 生成器状态内所有对象引用的扁平列表(GC mark 边)。
/// 生成器状态内全部引用边的扁平列表(GC mark 边):对象/字符串/BigInt 均产出,
/// 消费侧按值类型分发到对象栈与存活集。
pub(crate) fn generator_native_edges(obj: &JsObject) -> Vec<JsValue> {
let Some(state) = generator_state_mut(obj) else {
return Vec::new();
};
let mut edges = Vec::new();
let push = |v: JsValue, edges: &mut Vec<JsValue>| {
if v.is_object() {
edges.push(v);
}
};
push(state.callee, &mut edges);
edges.extend(state.args.iter().copied().filter(|v| v.is_object()));
push(state.result, &mut edges);
state.suspended.for_each_value(|v| {
if v.is_object() {
edges.push(v);
}
});
edges.push(state.callee);
edges.extend(state.args.iter().copied());
edges.push(state.result);
state.suspended.for_each_value(|v| edges.push(v));
edges
}
+9 -13
View File
@@ -1811,30 +1811,26 @@ fn promise_state_mut(obj: &JsObject) -> Option<&mut PromiseState> {
}
}
/// Promise 状态盒内所有对象引用(GC mark 边)。
/// Promise 状态盒内全部值边(GC mark 边):对象/字符串/BigInt 均产出,
/// 消费侧按值类型分发到对象栈与存活集。
pub(crate) fn promise_native_edges(obj: &JsObject) -> Vec<JsValue> {
let Some(state) = promise_state_ref(obj) else {
return Vec::new();
};
let mut edges = Vec::new();
let push = |v: JsValue, edges: &mut Vec<JsValue>| {
if v.is_object() {
edges.push(v);
}
};
push(state.result, &mut edges);
push(state.resolve_fn, &mut edges);
push(state.reject_fn, &mut edges);
edges.push(state.result);
edges.push(state.resolve_fn);
edges.push(state.reject_fn);
// 结算链指针:克隆恒为 session 表成员,凭这条边在原件/旧克隆仍存活时被
// 同轮置活,结算传导不会读到悬垂克隆。
if !state.promoted_clone.is_null() {
edges.push(JsValue::from_js_object(state.promoted_clone));
}
for r in &state.reactions {
push(r.promise, &mut edges);
push(r.resolve, &mut edges);
push(r.reject, &mut edges);
push(r.handler, &mut edges);
edges.push(r.promise);
edges.push(r.resolve);
edges.push(r.reject);
edges.push(r.handler);
}
edges
}
+118 -1
View File
@@ -1180,7 +1180,7 @@ mod tests {
use super::*;
use crate::vm::{CallFrame, FrameContinuation};
use oxide_builtins::{array_buffer, data_view, map, set, typed_array};
use oxide_builtins::{array_buffer, data_view, disposable_stack, map, set, typed_array};
use oxide_runtime_api::NativeResult;
fn plain_object(vm: &mut Vm) -> *mut JsObject {
@@ -1499,6 +1499,37 @@ mod tests {
val
}
/// 分配一个原型指向 DisposableStack.prototype 的占位对象并写入寄存器,
/// 作为构造器调用的 `this`。
fn dispose_stack_this(vm: &mut Vm, reg: u8) -> JsValue {
let proto = vm.session.builtin_world().disposable_stack_proto.as_ptr() as *mut JsObject;
let obj = vm.epoch.alloc(JsObject::new_empty(
oxide_kernel::shape_forge::EMPTY_SHAPE_ID,
JsValue::from_js_object(proto),
));
// 测试辅助函数绕过 alloc_object,需手动置位 EPOCH_BIT。
unsafe { (*obj).set_is_epoch(true) };
let val = JsValue::from_js_object(obj);
vm.regs[reg as usize] = val;
val
}
/// 置函数位标志的占位对象(通过 `is_callable` 判定),供 adopt 的
/// onDispose 槽使用;测试不调用它。
fn function_placeholder(vm: &mut Vm) -> JsValue {
let proto = vm.session.builtin_world().object_proto.as_ptr() as *mut JsObject;
let obj = vm.epoch.alloc(JsObject::new_empty(
oxide_kernel::shape_forge::EMPTY_SHAPE_ID,
JsValue::from_js_object(proto),
));
// 测试辅助函数绕过 alloc_object,需手动置位 EPOCH_BIT。
unsafe {
(*obj).set_is_epoch(true);
(*obj).set_function(true);
}
JsValue::from_js_object(obj)
}
#[test]
fn reset_maybe_collect_collects_after_threshold() {
let mut vm = vm_with_low_threshold();
@@ -1624,6 +1655,92 @@ mod tests {
assert_eq!(vm.gc_state.session_object_ptrs.len(), 2);
}
/// 盒持唯一引用的 session 串与 BigInt 经 Map native 边进入存活集:
/// 清寄存器仅留 Map 根后完整收集,两值仍登记在 session 表。
#[test]
fn session_gc_traces_map_string_and_bigint_value() {
let mut vm = vm_with_low_threshold();
map_this(&mut vm, 3);
let map_value = native_ok(map::map_constructor(&mut vm, &[3]));
let str = vm.new_string("map-box-string");
let str_ptr = str.as_string_ptr_mut();
let bi = vm.new_bigint(num_bigint::BigInt::from(121932631112635269u128));
let bi_ptr = bi.as_bigint_ptr() as *mut num_bigint::BigInt;
vm.regs[0] = map_value;
vm.regs[1] = JsValue::int(1);
vm.regs[2] = str;
native_ok(map::map_set(&mut vm, &[0, 1, 2]));
vm.regs[1] = JsValue::int(2);
vm.regs[2] = bi;
native_ok(map::map_set(&mut vm, &[0, 1, 2]));
let map_session = vm.promote_object(map_value.as_js_object_ptr());
vm.regs.fill(JsValue::undefined());
vm.regs[0] = JsValue::from_js_object(map_session);
collect(&mut vm);
// 只断言表成员——sweep 已释放的指针解引用即 UB。
assert!(vm.gc_state.session_string_ptrs.contains(&str_ptr));
assert!(vm.gc_state.session_bigint_ptrs.borrow().contains(&bi_ptr));
}
/// 盒持唯一引用的 session 串与 BigInt 经 Set native 边进入存活集:
/// 清寄存器仅留 Set 根后完整收集,两值仍登记在 session 表。
#[test]
fn session_gc_traces_set_string_and_bigint_value() {
let mut vm = vm_with_low_threshold();
set_this(&mut vm, 2);
let set_value = native_ok(set::set_constructor(&mut vm, &[2]));
let str = vm.new_string("set-box-string");
let str_ptr = str.as_string_ptr_mut();
let bi = vm.new_bigint(num_bigint::BigInt::from(987654321987654321u128));
let bi_ptr = bi.as_bigint_ptr() as *mut num_bigint::BigInt;
vm.regs[0] = set_value;
vm.regs[1] = str;
native_ok(set::set_add(&mut vm, &[0, 1]));
vm.regs[1] = bi;
native_ok(set::set_add(&mut vm, &[0, 1]));
let set_session = vm.promote_object(set_value.as_js_object_ptr());
vm.regs.fill(JsValue::undefined());
vm.regs[0] = JsValue::from_js_object(set_session);
collect(&mut vm);
// 只断言表成员——sweep 已释放的指针解引用即 UB。
assert!(vm.gc_state.session_string_ptrs.contains(&str_ptr));
assert!(vm.gc_state.session_bigint_ptrs.borrow().contains(&bi_ptr));
}
/// 盒持唯一引用的 session 串与 BigInt 经资源栈条目边进入存活集:
/// adopt 入栈后清寄存器仅留栈根,完整收集后两值仍登记在 session 表。
#[test]
fn session_gc_traces_dispose_stack_string_and_bigint_value() {
let mut vm = vm_with_low_threshold();
dispose_stack_this(&mut vm, 3);
let stack_value = native_ok(disposable_stack::disposable_stack_constructor(&mut vm, &[3]));
let on_dispose = function_placeholder(&mut vm);
let str = vm.new_string("stack-box-string");
let str_ptr = str.as_string_ptr_mut();
let bi = vm.new_bigint(num_bigint::BigInt::from(4611686018427387904u128));
let bi_ptr = bi.as_bigint_ptr() as *mut num_bigint::BigInt;
vm.regs[0] = stack_value;
vm.regs[1] = str;
vm.regs[2] = on_dispose;
native_ok(disposable_stack::disposable_stack_adopt(&mut vm, &[0, 1, 2]));
vm.regs[1] = bi;
native_ok(disposable_stack::disposable_stack_adopt(&mut vm, &[0, 1, 2]));
let stack_session = vm.promote_object(stack_value.as_js_object_ptr());
vm.regs.fill(JsValue::undefined());
vm.regs[0] = JsValue::from_js_object(stack_session);
collect(&mut vm);
// 只断言表成员——sweep 已释放的指针解引用即 UB。
assert!(vm.gc_state.session_string_ptrs.contains(&str_ptr));
assert!(vm.gc_state.session_bigint_ptrs.borrow().contains(&bi_ptr));
}
#[test]
fn session_gc_keeps_shared_array_buffer_alive_through_view_native_edges() {
let mut vm = vm_with_low_threshold();
+186
View File
@@ -335,3 +335,189 @@ fn dead_closure_upvalues_freed_by_sweep_without_double_free() {
freed_with_captures - freed_no_captures
);
}
// ── 原生盒内字符串/BigInt 边存活(mark 边收集去对象预过滤) ─────────────────
/// 盒持唯一引用 session 串(Promise 结算值):churn 窗口内执行期收集
/// 仅能经状态盒边闭合存活,结算交付后读回精确值。churn 串与盒串同长:
/// 若 box 释放后其内存被同长分配复用,读回即错值。
#[test]
fn promise_result_string_survives_runtime_gc() {
let mut vm = vm_with_threshold(512);
let first = compile(
"(function(){ \
var s = 'prombox'.repeat(8); \
var p = Promise.resolve(s); \
p.then(function(v){ globalThis.got = v; }); \
globalThis.p = p; })(); \
for (var i = 0; i < 2000; i++) { var t = 'z'.repeat(56); } \
0",
);
vm.run(&Arc::new(first)).expect("run1");
let second = compile("globalThis.got");
let result = vm.run(&Arc::new(second)).expect("run2");
let text = vm.lookup_str(result).expect("got 应为字符串").to_string();
assert_eq!(text, "prombox".repeat(8));
assert!(vm.session_gc_stats().total_collections > 0, "churn 应触发执行期字符串 GC");
}
/// 盒持唯一引用 session BigInt(Promise 结算值,运行时乘积非常量池字面量):
/// 同窗口存活闭合经 live_bigints,结算交付后读回精确值。
#[test]
fn promise_result_bigint_survives_runtime_gc() {
let mut vm = vm_with_threshold(512);
let first = compile(
"(function(){ \
var b = 987654321n * 123456789n; \
var p = Promise.resolve(b); \
p.then(function(v){ globalThis.got = v; }); \
globalThis.p = p; })(); \
for (var i = 0; i < 2000; i++) { var t = 'churn' + i; } \
0",
);
vm.run(&Arc::new(first)).expect("run1");
let second = compile("globalThis.got");
let result = vm.run(&Arc::new(second)).expect("run2");
let expected = num_bigint::BigInt::from(987654321u64) * num_bigint::BigInt::from(123456789u64);
assert!(result.is_bigint(), "got 应为 BigInt");
assert_eq!(vm.bigint_value(result), &expected);
assert!(vm.session_gc_stats().total_collections > 0, "churn 应触发执行期 GC");
}
/// 挂起生成器帧寄存器持唯一引用 session BigInt:yield 挂起后仅状态盒
/// 可达,churn 窗口收集闭合存活,恢复读回精确值。
#[test]
fn suspended_generator_bigint_survives_runtime_gc() {
let mut vm = vm_with_threshold(512);
let module = compile(
"(function(){ \
function* gen() { var b = 987654321n * 123456789n; yield 1; return b; } \
var g = gen(); g.next(); globalThis.g = g; })(); \
for (var i = 0; i < 2000; i++) { var t = 'churn' + i; } \
globalThis.g.next().value",
);
let result = vm.run(&Arc::new(module)).expect("run");
let expected = num_bigint::BigInt::from(987654321u64) * num_bigint::BigInt::from(123456789u64);
assert!(result.is_bigint(), "恢复返回值应为 BigInt");
assert_eq!(vm.bigint_value(result), &expected);
assert!(vm.session_gc_stats().total_collections > 0, "churn 应触发执行期 GC");
}
/// 挂起异步函数帧持唯一引用 session BigInt:await 挂起后仅状态盒可达,
/// 微任务恢复结算后读回精确值。
#[test]
fn suspended_async_function_bigint_survives_runtime_gc() {
let mut vm = vm_with_threshold(512);
let first = compile(
"(function(){ \
async function f() { var b = 987654321n * 123456789n; await 0; return b; } \
var p = f(); \
p.then(function(v){ globalThis.got = v; }); \
globalThis.p = p; })(); \
for (var i = 0; i < 2000; i++) { var t = 'churn' + i; } \
0",
);
vm.run(&Arc::new(first)).expect("run1");
let second = compile("globalThis.got");
let result = vm.run(&Arc::new(second)).expect("run2");
let expected = num_bigint::BigInt::from(987654321u64) * num_bigint::BigInt::from(123456789u64);
assert!(result.is_bigint(), "got 应为 BigInt");
assert_eq!(vm.bigint_value(result), &expected);
assert!(vm.session_gc_stats().total_collections > 0, "churn 应触发执行期 GC");
}
/// 挂起异步生成器帧持唯一引用 session BigInt:首次 next 挂起后仅状态盒
/// 可达,挂起期 churn 窗口的收集闭合存活,同 run 内恢复并交付精确值。
#[test]
fn suspended_async_generator_bigint_survives_runtime_gc() {
let mut vm = vm_with_threshold(512);
let first = compile(
"async function* gen() { var b = 987654321n * 123456789n; yield 1; return b; } \
async function run() { \
var it = gen(); globalThis.g = it; \
var p1 = it.next(); \
for (var i = 0; i < 2000; i++) { var t = 'churn' + i; } \
await p1; \
return (await it.next()).value; } \
run().then(function(v){ globalThis.got = v; })",
);
vm.run(&Arc::new(first)).expect("run1");
let second = compile("globalThis.got");
let result = vm.run(&Arc::new(second)).expect("run2");
let expected = num_bigint::BigInt::from(987654321u64) * num_bigint::BigInt::from(123456789u64);
assert!(result.is_bigint(), "got 应为 BigInt");
assert_eq!(vm.bigint_value(result), &expected);
assert!(vm.session_gc_stats().total_collections > 0, "churn 应触发执行期 GC");
}
/// 盒持唯一引用 session 串(Map 值):churn 窗口内收集经盒边闭合存活,
/// get 读回精确值。churn 串与盒串同长:若 box 释放后其内存被同长
/// 分配复用,读回即错值。
#[test]
fn map_value_string_survives_runtime_gc() {
let mut vm = vm_with_threshold(512);
let module = compile(
"(function(){ \
var s = 'mapbox'.repeat(8); \
var m = new Map(); m.set('k', s); globalThis.m = m; })(); \
for (var i = 0; i < 2000; i++) { var t = 'z'.repeat(48); } \
globalThis.m.get('k')",
);
let result = vm.run(&Arc::new(module)).expect("run");
let text = vm.lookup_str(result).expect("map 值应为字符串").to_string();
assert_eq!(text, "mapbox".repeat(8));
assert!(vm.session_gc_stats().total_collections > 0, "churn 应触发执行期字符串 GC");
}
/// 盒持唯一引用 session 串(Set 元素):churn 窗口内收集经盒边闭合存活,
/// values 迭代读回精确值。churn 串与盒串同长:若 box 释放后其内存被
/// 同长分配复用,读回即错值。
#[test]
fn set_value_string_survives_runtime_gc() {
let mut vm = vm_with_threshold(512);
let module = compile(
"(function(){ \
var s = 'setbox'.repeat(8); \
var st = new Set(); st.add(s); globalThis.st = st; })(); \
for (var i = 0; i < 2000; i++) { var t = 'z'.repeat(48); } \
globalThis.st.values().next().value",
);
let result = vm.run(&Arc::new(module)).expect("run");
let text = vm.lookup_str(result).expect("set 元素应为字符串").to_string();
assert_eq!(text, "setbox".repeat(8));
assert!(vm.session_gc_stats().total_collections > 0, "churn 应触发执行期字符串 GC");
}
/// 盒持唯一引用 session 串(DisposableStack adopt 值):churn 窗口内收集
/// 经条目边闭合存活,dispose 回调交付后读回精确值。churn 串与盒串同长:
/// 若 box 释放后其内存被同长分配复用,读回即错值。
#[test]
fn disposable_stack_value_string_survives_runtime_gc() {
let mut vm = vm_with_threshold(512);
let module = compile(
"(function(){ \
var s = 'stackbox'.repeat(8); \
var st = new DisposableStack(); \
st.adopt(s, function(v){ globalThis.got = v; }); \
globalThis.st = st; })(); \
for (var i = 0; i < 2000; i++) { var t = 'z'.repeat(64); } \
globalThis.st.dispose(); \
globalThis.got",
);
let result = vm.run(&Arc::new(module)).expect("run");
let text = vm.lookup_str(result).expect("adopt 值应为字符串").to_string();
assert_eq!(text, "stackbox".repeat(8));
assert!(vm.session_gc_stats().total_collections > 0, "churn 应触发执行期字符串 GC");
}