策划panic时把代码位置和负载写入design_debug
Project CI / AI game creator shell Rust shard 1/4 (pull_request) Has been cancelled
Project CI / AI game creator shell Rust shard 2/4 (pull_request) Has been cancelled
Project CI / AI game creator shell Rust shard 3/4 (pull_request) Has been cancelled
Project CI / AI game creator shell Rust shard 4/4 (pull_request) Has been cancelled
Project CI / AI game creator shell Rust smoke (pull_request) Has been cancelled
Project CI / AI game creator shell Rust crates (pull_request) Has been cancelled
Project CI / Backend tests (pull_request) Has been cancelled
Project CI / Native shell tests (pull_request) Has been cancelled
Project CI / Frontend tests (pull_request) Has been cancelled
Project CI / Repository checks (pull_request) Has been cancelled
Project CI / AI game creator shell web tests (pull_request) Has been cancelled

运行段经tokio task-local携带项目根,多线程runtime跨worker迁移也能归因
首次运行安装panic hook(take_hook包装原hook,update_hook在1.96仍unstable)
hook仅在策划运行段内生效,其它任务直接透传原hook
design_panic_error只返回固定公开文案,debug记录由hook统一写入
回归测试断言panic记录落盘且包含design_runtime.rs位置与注入payload
This commit is contained in:
2026-09-17 09:27:48 +00:00
parent 97bf8cb864
commit 73729394e2
2 changed files with 74 additions and 13 deletions
@@ -15,6 +15,43 @@ const DESIGN_ACTIVE_LOCK: &str = ".agent/design-agent/active.lock";
const DESIGN_PANIC_PUBLIC_ERROR: &str =
"策划运行发生内部错误,本轮已中断,可直接重试;若反复出现请反馈。";
// 运行段经 task-local 携带项目根,panic hook 据此把位置和负载写进私有 design_debug。
// task-local 而非 thread-local:多线程 runtime 下 future 会跨 worker 迁移。
tokio::task_local! {
static DESIGN_PANIC_ROOT: Option<PathBuf>;
}
fn ensure_design_panic_hook() {
static ONCE: std::sync::Once = std::sync::Once::new();
ONCE.call_once(|| {
let previous = std::panic::take_hook();
std::panic::set_hook(Box::new(move |info| {
if let Ok(Some(root)) = DESIGN_PANIC_ROOT.try_with(Clone::clone) {
let location = info
.location()
.map(|location| {
format!(
"{}:{}:{}",
location.file(),
location.line(),
location.column()
)
})
.unwrap_or_else(|| "未知位置".to_string());
design_debug(
&root,
"panic",
json!({
"location": location,
"error": info.payload_as_str().unwrap_or("未知 panic 负载"),
}),
);
}
previous(info);
}));
});
}
#[derive(Clone, Debug, Deserialize, Serialize)]
#[serde(
tag = "type",
@@ -982,13 +1019,15 @@ async fn finish_design_command(
// panic 边界:运行期 panic 转成普通失败,交给既有错误分支恢复(重读检查点、
// 写 last_error、发最终 view)。否则 unwind 会杀死 command taskIPC 永不
// 返回(前端停在工作态),会话停在无错误的 pending,用户只能看到无声的重试。
let outcome =
std::panic::AssertUnwindSafe(run_design_loop(root, resources, &mut session, &mut emit))
.catch_unwind()
.await;
ensure_design_panic_hook();
let run = DESIGN_PANIC_ROOT.scope(
Some(root.to_path_buf()),
run_design_loop(root, resources, &mut session, &mut emit),
);
let outcome = std::panic::AssertUnwindSafe(run).catch_unwind().await;
let result = match outcome {
Ok(result) => result,
Err(payload) => Err(design_panic_error(root, payload)),
Err(payload) => Err(design_panic_error(payload)),
};
if let Err(error) = result {
// 从最后一个持久检查点恢复,防止写后未记结果被误认为已完成。
@@ -1010,14 +1049,9 @@ async fn finish_design_command(
Ok(view)
}
// panic 负载可能包含路径或内容片段,公开文案固定;细节只进私有 design_debug。
fn design_panic_error(root: &Path, payload: Box<dyn std::any::Any + Send>) -> String {
let detail = payload
.downcast_ref::<&str>()
.map(|value| (*value).to_string())
.or_else(|| payload.downcast_ref::<String>().cloned())
.unwrap_or_else(|| "未知 panic 负载".to_string());
design_debug(root, "panic", json!({"error": detail}));
// panic 负载可能包含路径或内容片段,公开文案固定;位置和负载由 panic hook 写进私有
// design_debugtask-local 提供项目根),不进入用户可见消息。
fn design_panic_error(_payload: Box<dyn std::any::Any + Send>) -> String {
DESIGN_PANIC_PUBLIC_ERROR.to_string()
}
@@ -1665,6 +1699,7 @@ mod tests {
#[tokio::test(flavor = "current_thread")]
async fn design_tool_panic_becomes_visible_retryable_error() {
let (_temp, root, resources) = init_design_project();
std::env::set_var("GENARRATIVE_AGC_DESIGN_DEBUG", "1");
let panic_call = platform_llm::LlmToolCall {
id: "call-panic".into(),
name: "design_test__panic".into(),
@@ -1705,6 +1740,30 @@ mod tests {
"panic 不得写半个工具输出"
);
// design_debug 经独立线程落盘,轮询等待 panic 记录出现。
let debug_dir = root.join(".debug/design-agent");
let mut panic_record = None;
for _ in 0..100 {
panic_record = fs::read_dir(&debug_dir).ok().and_then(|entries| {
entries.flatten().find_map(|entry| {
let path = entry.path();
let name = path.file_name()?.to_string_lossy().into_owned();
if name.ends_with("-panic.json") {
fs::read_to_string(path).ok()
} else {
None
}
})
});
if panic_record.is_some() {
break;
}
tokio::time::sleep(Duration::from_millis(20)).await;
}
let panic_record = panic_record.expect("panic hook 必须把位置和负载写入 design_debug");
assert!(panic_record.contains("design_runtime.rs"));
assert!(panic_record.contains("注入的策划工具 panic"));
let view =
continue_design_agent_at(&root, &resources, "turn-retry", DesignInput::Retry, |_| {})
.await
@@ -8,6 +8,8 @@
`finish_design_command` 的运行段包在 `catch_unwind` 边界内:任何运行期 panic 被转成一次普通失败,由既有错误分支从最后一个持久检查点恢复,向会话写入固定公开文案"策划运行发生内部错误,本轮已中断,可直接重试;若反复出现请反馈。",并以 `running=false、可重试、lastError 有值` 的视图正常返回。panic 负载只写入私有 design_debug,不进入用户可见消息;continue、recover_uncertain 与 decide 三个入口共享同一边界。边界不改变工具错误、Provider 瞬态重试和批次不确定恢复的既有语义。
运行段通过 task-local 携带项目根;首次运行时安装的 panic hook 在 panic 瞬间把代码位置(file:line:column)和负载追加写入私有 design_debug,随后交给原 hook 维持既有 stderr 输出。hook 只在策划运行段内生效(其它任务无 task-local 上下文时直接透传),多线程 runtime 下 future 跨 worker 迁移也能正确归因。
## 资源画布交互与工作台状态同步
- 工作台向窗口标题栏发布正在运行的项目时,输入未变化不得形成重复发布与清理的渲染循环;打开项目动作始终使用当前工作台处理逻辑,退出工作台后清除其标题栏状态。