Update: 将子项目从 submodule 转为完整内容

- 移除 GovAI, nomifun-tauri, 算力盒子的 submodule 引用 - 添加所有子项目的完整源代码 - 保留原始 .git 为 .git.bak 备份
2026-07-04 19:20:46 +08:00
parent 54d6465fa7
commit f7a720204a
3360 changed files with 802660 additions and 3 deletions
@@ -0,0 +1,327 @@
+//! observe 注入契约快照测试（`#[ignore]`，本机/打包 chrome）。
+//!
+//! 对一个**固定 HTML fixture**（`tests/fixtures/iframe.html`，file:// 加载）跑注入侧的
+//! `incrementalAriaSnapshot`，把它返回的 `.full`（aria YAML，给 LLM 看的那一版）冻成 insta
+//! 快照。目的：
+//! - **供人审**：aria 输出形态（role/name/ref=f0e<n>）一眼可读、可在 review 里核对；
+//! - **防漂移**：vendor 的 Playwright InjectedScript 升级后若 aria 序列化形态变了，快照 diff
+//!   会立刻报出来（DESIGN：整包 vendor 不 fork，靠契约测试钉住外部行为）。
+//!
+//! 接线母本 = `src/injected.rs` 的 `inject_aria_snapshot_smoke`（八步：launch → connect →
+//! run_attach_loop → enable_auto_attach → createTarget page → navigate(file://) → arm →
+//! `Runtime.evaluate("document.body")` 取 objectId → `call_injected("incrementalAriaSnapshot",
+//! [body, opts], by_value=true)` → 取 `.full`）。
+//!
+//! 手动跑（本机 Windows 有系统 Chrome）：
+//!   set NOMIFUN_CHROME_BINARY=...\chrome.exe
+//!   cargo nextest run -p nomi-browser-engine --run-ignored all -E 'test(observe_inject_contract)'
+//! 首跑写 `.snap.new`；`cargo insta accept`（或手动改名 .snap）接受为基线。
+//! 跑完核对任务管理器无残留 chrome（Builder kill_on_drop 应自动清）。
+
+use std::time::Duration;
+
+use chromiumoxide::cdp::browser_protocol::page::{
+    EnableParams as PageEnable, NavigateParams,
+};
+use chromiumoxide::cdp::browser_protocol::target::{CreateTargetParams, EventAttachedToTarget};
+use chromiumoxide::cdp::js_protocol::runtime::{
+    CallArgument, EvaluateParams, ExecutionContextId, RemoteObjectId,
+};
+
+use nomi_browser_engine::injected::InjectionManager;
+use nomi_browser_engine::launch::{launch_chrome, LaunchConfig};
+use nomi_browser_engine::transport::Connection;
+
+mod common;
+
+/// fixture 的 file:// URL（`file://` + 单前导斜杠 + POSIX 路径；unix `/abs` 已带斜杠、windows
+/// `C:/abs` 补一个。旧 `file:///{manifest}` 在 unix 产四斜杠，被 chrome 归一后破坏 url 比较）。
+fn fixture_url() -> String {
+    let manifest = env!("CARGO_MANIFEST_DIR").replace('\\', "/");
+    let abs = if manifest.starts_with('/') {
+        manifest
+    } else {
+        format!("/{manifest}")
+    };
+    format!("file://{abs}/tests/fixtures/iframe.html")
+}
+
+#[tokio::test]
+#[ignore = "需本机/打包 chrome：set NOMIFUN_CHROME_BINARY 后 --run-ignored all"]
+async fn observe_inject_contract_iframe() {
+    // 1) resolve chrome（env NOMIFUN_CHROME_BINARY > 打包 > 数据目录 > 下载兜底）+ launch headless。
+    let chrome = nomi_browser_engine::acquire::resolve_chrome_path(
+        &std::env::temp_dir().join("nomifun-browser-data"),
+        None,
+    )
+    .await
+    .expect("resolve chrome (set NOMIFUN_CHROME_BINARY)");
+    let cfg = LaunchConfig {
+        chrome_path: chrome,
+        user_data_dir: std::env::temp_dir().join("nomifun-observe-contract-profile"),
+        headful: false,
+    };
+    let launched = launch_chrome(&cfg, true).await.expect("launch chrome");
+
+    // 2) connect + 先 run_attach_loop 后 enable_auto_attach（顺序铁律，否则首批 attach 事件丢）。
+    let _child = launched.child; // 保活 chrome（drop 即清理）。
+    let conn = Connection::connect_launched(launched.transport)
+        .await
+        .expect("connect");
+    let _attach_loop = conn.run_attach_loop();
+    conn.enable_auto_attach().await.expect("auto attach");
+
+    // 3) 取一个 page session（createTarget about:blank + 等其 attachedToTarget）。
+    let mut attached = conn.subscribe(EventAttachedToTarget::IDENTIFIER, None);
+    let create = CreateTargetParams::new("about:blank");
+    let cr = conn
+        .send::<CreateTargetParams>(nomi_browser_engine::transport::ROOT_SESSION, &create)
+        .await
+        .expect("createTarget");
+    let target_id = cr["targetId"].as_str().expect("targetId").to_string();
+    let page_session = loop {
+        let ev = tokio::time::timeout(Duration::from_secs(10), attached.recv())
+            .await
+            .expect("attach timeout")
+            .expect("attach recv");
+        if let Ok(att) = serde_json::from_value::<EventAttachedToTarget>(ev.params.clone()) {
+            let tid: String = att.target_info.target_id.clone().into();
+            if tid == target_id && att.target_info.r#type == "page" {
+                break String::from(att.session_id);
+            }
+        }
+    };
+
+    // 4) navigate 到固定 fixture（file://，含 h1/button/textbox/iframe，aria 形态稳定）。
+    conn.send::<PageEnable>(&page_session, &PageEnable::default())
+        .await
+        .expect("Page.enable");
+    let mut load_rx = conn.subscribe("Page.loadEventFired", Some(&page_session));
+    conn.send::<NavigateParams>(&page_session, &NavigateParams::new(fixture_url()))
+        .await
+        .expect("navigate");
+    let _ = tokio::time::timeout(Duration::from_secs(15), load_rx.recv()).await;
+
+    // 5) arm 注入管线 + 等主 frame 的 utility-world context 就绪（主 frameId = page targetId）。
+    let mgr = InjectionManager::new(conn.clone(), page_session.clone());
+    let _ctx_loop = mgr.arm().await.expect("arm injection");
+    let frame_id = target_id.clone();
+    let mut ctx_id = None;
+    for _ in 0..50 {
+        if let Ok(id) = mgr.context_id_for(&frame_id) {
+            ctx_id = Some(id);
+            break;
+        }
+        tokio::time::sleep(Duration::from_millis(100)).await;
+    }
+    let ctx_id = ctx_id.expect("utility world context never registered for main frame");
+
+    // 6) 取 document.body 的 objectId（同 utility world 的元素句柄），作 incrementalAriaSnapshot 的 node。
+    let mut body_eval = EvaluateParams::new("document.body".to_string());
+    body_eval.context_id = Some(ExecutionContextId::new(ctx_id));
+    body_eval.return_by_value = Some(false);
+    let body_res = conn
+        .send::<EvaluateParams>(&page_session, &body_eval)
+        .await
+        .expect("evaluate document.body");
+    let body_obj_id = body_res["result"]["objectId"]
+        .as_str()
+        .expect("document.body objectId")
+        .to_string();
+
+    // 7) call_injected incrementalAriaSnapshot(body, {mode:ai, refPrefix:f0, depth:12})；node 走
+    //    objectId，opts 走 by-value；返回 by-value（result.value 是 {full, iframeRefs, ...}）。
+    let node_arg = CallArgument {
+        object_id: Some(RemoteObjectId::new(body_obj_id)),
+        ..Default::default()
+    };
+    let opts_arg = CallArgument {
+        value: Some(serde_json::json!({"mode": "ai", "refPrefix": "f0", "depth": 12})),
+        ..Default::default()
+    };
+    let result = mgr
+        .call_injected(
+            &frame_id,
+            "incrementalAriaSnapshot",
+            vec![node_arg, opts_arg],
+            true,
+        )
+        .await
+        .expect("call_injected incrementalAriaSnapshot");
+
+    // 8) 取 .full（aria YAML 字符串）并快照。aria YAML 只含页面结构（generic/button/textbox/iframe
+    //    + ref=f0e<n>），不含随机 world 名，故无需归一。
+    let full = result["value"]["full"]
+        .as_str()
+        .expect("incrementalAriaSnapshot result.value.full is a string")
+        .to_string();
+    assert!(!full.trim().is_empty(), "aria .full must be non-empty");
+
+    insta::assert_snapshot!(full);
+}
+
+// ═══════════════════════════════════════════════════════════════════════════
+// 任务 6：CdpBackend::observe 全链集成测试（`#[ignore]`，本机/打包 chrome）。
+// 复用 tests/common 的 build_backend_for_fixture helper（勿再复制契约母本）。
+// 被测对象是 engine.observe()：逐帧 incrementalAriaSnapshot → 缝合 → 脱敏 → 代际翻新 ref 表。
+// ═══════════════════════════════════════════════════════════════════════════
+
+use nomi_browser_engine::{BrowserEngine, ObserveOpts};
+
+/// 同进程 iframe 缝合：navigate iframe.html（srcdoc iframe 是同进程，应能缝）→ observe →
+/// 快照 yaml + 断言含父帧 ref f0e*；若同进程子帧缝上则含子内容（缩进的 Inner 链接）。
+#[tokio::test]
+#[ignore = "需本机/打包 chrome：set NOMIFUN_CHROME_BINARY 后 --run-ignored all"]
+async fn observe_iframe_stitched() {
+    let engine = common::build_backend_for_fixture("iframe").await;
+    engine
+        .navigate(&common::fixture_url("iframe.html"), false)
+        .await
+        .expect("navigate iframe.html");
+    let obs = engine
+        .observe(&ObserveOpts::default())
+        .await
+        .expect("observe");
+
+    eprintln!("=== observe_iframe_stitched yaml ===\n{}\n=== end ===", obs.yaml);
+    // 父帧 ref（f0e*）必现。
+    assert!(obs.yaml.contains("f0e"), "expected parent-frame ref f0e*:\n{}", obs.yaml);
+    // <data> 包裹。
+    assert!(obs.yaml.contains("<data"), "expected <data> wrap:\n{}", obs.yaml);
+    // 同进程 srcdoc 子帧应缝入（含其链接文本 Inner）；srcdoc iframe 是同进程，应能缝。
+    //
+    // 快照里的 origin 是本机绝对 file:// 路径——快照前归一成稳定占位符，使快照可移植
+    // （不钉死 checkout 路径），同时仍审 <data origin=…> 包裹 + 缝合结构。
+    let normalized = normalize_fixture_origin(&obs.yaml, "iframe.html");
+    insta::assert_snapshot!("observe_iframe_stitched", normalized);
+}
+
+/// 把 yaml 里本机绝对 `origin="file:///.../tests/fixtures/<name>"` 归一成
+/// `origin="file://<FIXTURE>/<name>"`，让 insta 快照不钉死 checkout 路径。
+fn normalize_fixture_origin(yaml: &str, name: &str) -> String {
+    let manifest = env!("CARGO_MANIFEST_DIR").replace('\\', "/");
+    // Chrome reports file:// origins with exactly ONE leading slash before the
+    // POSIX path (unix: `file:///abs`; windows: `file:///C:/abs`). But
+    // CARGO_MANIFEST_DIR is `/abs` on unix and `C:/abs` on windows — so prepend a
+    // slash only when it doesn't already start with one. The old unconditional
+    // `file:///{manifest}` produced FOUR slashes on unix (`file:////Users/...`),
+    // never matched the reported origin, and leaked the absolute checkout path
+    // into the snapshot. This makes the placeholder fire on every host.
+    let abs = if manifest.starts_with('/') {
+        manifest
+    } else {
+        format!("/{manifest}")
+    };
+    let actual = format!("file://{abs}/tests/fixtures/{name}");
+    let placeholder = format!("file://<FIXTURE>/{name}");
+    yaml.replace(&actual, &placeholder)
+}
+
+/// open shadow 可见、closed shadow 不可见（D11）。
+#[tokio::test]
+#[ignore = "需本机/打包 chrome：set NOMIFUN_CHROME_BINARY 后 --run-ignored all"]
+async fn observe_shadow_open_visible_closed_absent() {
+    let engine = common::build_backend_for_fixture("shadow").await;
+    engine
+        .navigate(&common::fixture_url("shadow.html"), false)
+        .await
+        .expect("navigate shadow.html");
+    let obs = engine
+        .observe(&ObserveOpts::default())
+        .await
+        .expect("observe");
+
+    eprintln!("=== observe_shadow yaml ===\n{}\n=== end ===", obs.yaml);
+    assert!(
+        obs.yaml.contains("OpenShadowBtn"),
+        "open shadow content must be visible:\n{}",
+        obs.yaml
+    );
+    assert!(
+        !obs.yaml.contains("ClosedShadowBtn"),
+        "closed shadow content must NOT be visible:\n{}",
+        obs.yaml
+    );
+}
+
+/// 代际单调递增（无导航两次 observe）+ 导航后旧 ref 在新表 resolve 不到（generation 跳变隔离）。
+#[tokio::test]
+#[ignore = "需本机/打包 chrome：set NOMIFUN_CHROME_BINARY 后 --run-ignored all"]
+async fn observe_generation_increments_and_renews() {
+    let engine = common::build_backend_for_fixture("gen").await;
+    engine
+        .navigate(&common::fixture_url("shadow.html"), false)
+        .await
+        .expect("navigate");
+    let obs1 = engine.observe(&ObserveOpts::default()).await.expect("observe1");
+    let obs2 = engine.observe(&ObserveOpts::default()).await.expect("observe2");
+    // 无导航两次 observe：代际单调 +1。
+    assert_eq!(
+        obs2.generation.0,
+        obs1.generation.0 + 1,
+        "generation must increment per observe: {:?} -> {:?}",
+        obs1.generation,
+        obs2.generation
+    );
+    // 导航到不同页再 observe：代际继续递增（旧代际 ref 不再属于新表）。
+    engine
+        .navigate(&common::fixture_url("secrets.html"), false)
+        .await
+        .expect("navigate2");
+    let obs3 = engine.observe(&ObserveOpts::default()).await.expect("observe3");
+    assert!(
+        obs3.generation.0 > obs2.generation.0,
+        "generation must advance after navigation: {:?} -> {:?}",
+        obs2.generation,
+        obs3.generation
+    );
+    // 每代 entries 应非空（页面有可操作元素：button 等）。
+    assert!(!obs1.entries.is_empty(), "obs1 entries should be non-empty");
+    assert!(!obs3.entries.is_empty(), "obs3 entries should be non-empty");
+}
+
+/// 脱敏：secrets.html 的明文 sk-/Bearer token 不出现在 yaml，且整体被 <data> 包裹。
+#[tokio::test]
+#[ignore = "需本机/打包 chrome：set NOMIFUN_CHROME_BINARY 后 --run-ignored all"]
+async fn observe_redacts_secrets() {
+    let engine = common::build_backend_for_fixture("secrets").await;
+    engine
+        .navigate(&common::fixture_url("secrets.html"), false)
+        .await
+        .expect("navigate secrets.html");
+    let obs = engine
+        .observe(&ObserveOpts::default())
+        .await
+        .expect("observe");
+
+    eprintln!("=== observe_redacts_secrets yaml ===\n{}\n=== end ===", obs.yaml);
+    assert!(obs.yaml.contains("<data"), "expected <data> wrap:\n{}", obs.yaml);
+    // 明文 secret 实体不得出现。
+    assert!(
+        !obs.yaml.contains("sk-ABCDEFGHIJ0123456789xyzQRSTUV"),
+        "plaintext sk- token leaked:\n{}",
+        obs.yaml
+    );
+    assert!(
+        !obs.yaml.contains("abcdef0123456789ABCDEFghij"),
+        "plaintext Bearer token leaked:\n{}",
+        obs.yaml
+    );
+    // D5（Critical）：password <input value="hun]ter2sk"> 的明文 value 不得进 YAML。
+    // 序列化层（observe）按 DOM type=password 收 ref 后宿主侧抹掉内联 value（短/低熵口令
+    // 正则脱敏救不了，必须靠 type=password 信号置空）。**value 含 `]`** 压测确定性锚点 bug：
+    // 旧 blank_inline_value 用全行 rfind(']') → 锚点落进 value 内部漏抹 → 明文泄露；改 ref-token
+    // 锚后必不泄露。
+    assert!(
+        !obs.yaml.contains("hun]ter2"),
+        "password value (with ]) leaked into observe yaml:\n{}",
+        obs.yaml
+    );
+    // 脱敏占位符应在（确认确实脱敏，而非该文本压根没进 aria 树）。
+    assert!(
+        obs.yaml.contains("[REDACTED_SECRET]"),
+        "expected redaction placeholder (text must have entered the aria tree):\n{}",
+        obs.yaml
+    );
+}
+