Merge remote-tracking branch 'origin/main' into pr471-push

# Conflicts:
#	crates/obscura-js/js/bootstrap.js

build.log

@@ -0,0 +1,283 @@
+   Compiling proc-macro2 v1.0.106
+   Compiling quote v1.0.45
+   Compiling unicode-ident v1.0.24
+   Compiling libc v0.2.186
+   Compiling cfg-if v1.0.4
+   Compiling parking_lot_core v0.9.12
+   Compiling serde_core v1.0.228
+   Compiling serde v1.0.228
+   Compiling smallvec v1.15.2
+   Compiling icu_properties_data v2.2.0
+   Compiling icu_normalizer_data v2.2.0
+   Compiling stable_deref_trait v1.2.1
+   Compiling pin-project-lite v0.2.17
+   Compiling version_check v0.9.5
+   Compiling scopeguard v1.2.0
+   Compiling itoa v1.0.18
+   Compiling memchr v2.8.2
+   Compiling typenum v1.20.1
+   Compiling lock_api v0.4.14
+   Compiling bytes v1.11.1
+   Compiling crc32fast v1.5.0
+   Compiling zmij v1.0.21
+   Compiling generic-array v0.14.7
+   Compiling rand_core v0.6.4
+   Compiling thiserror v2.0.18
+   Compiling rand v0.8.6
+   Compiling serde_json v1.0.150
+   Compiling subtle v2.6.1
+   Compiling futures-core v0.3.32
+   Compiling syn v2.0.117
+   Compiling glob v0.3.3
+   Compiling litemap v0.8.2
+   Compiling writeable v0.6.3
+   Compiling utf8_iter v1.0.4
+   Compiling prettyplease v0.2.37
+   Compiling clang-sys v1.8.1
+   Compiling siphasher v1.0.3
+   Compiling getrandom v0.2.17
+   Compiling errno v0.3.14
+   Compiling mio v1.2.1
+   Compiling socket2 v0.6.4
+   Compiling signal-hook-registry v1.4.8
+   Compiling autocfg v1.5.1
+   Compiling parking_lot v0.12.5
+   Compiling futures-sink v0.3.32
+   Compiling phf_shared v0.11.3
+   Compiling crypto-common v0.1.7
+   Compiling rustix v0.38.44
+   Compiling regex-syntax v0.8.11
+   Compiling percent-encoding v2.3.2
+   Compiling num-traits v0.2.19
+   Compiling either v1.16.0
+   Compiling bitflags v2.13.0
+   Compiling minimal-lexical v0.2.1
+   Compiling phf_generator v0.11.3
+   Compiling futures-channel v0.3.32
+   Compiling nom v7.1.3
+   Compiling libloading v0.8.9
+   Compiling linux-raw-sys v0.4.15
+   Compiling bindgen v0.71.1
+   Compiling rustc-hash v2.1.2
+   Compiling itertools v0.13.0
+   Compiling log v0.4.32
+   Compiling futures-io v0.3.32
+   Compiling shlex v1.3.0
+   Compiling adler2 v2.0.1
+   Compiling home v0.5.12
+   Compiling once_cell v1.21.4
+   Compiling hashbrown v0.17.1
+   Compiling regex-automata v0.4.14
+   Compiling equivalent v1.0.2
+   Compiling radium v0.7.0
+   Compiling slab v0.4.12
+   Compiling futures-task v0.3.32
+   Compiling indexmap v2.14.0
+   Compiling cexpr v0.6.0
+   Compiling miniz_oxide v0.8.9
+   Compiling gzip-header v1.1.0
+   Compiling form_urlencoded v1.2.2
+   Compiling fslock v0.2.1
+   Compiling find-msvc-tools v0.1.9
+   Compiling shlex v2.0.1
+   Compiling cc v1.2.64
+   Compiling which v6.0.3
+   Compiling http v1.4.2
+   Compiling paste v1.0.15
+   Compiling synstructure v0.13.2
+   Compiling regex v1.12.4
+   Compiling heck v0.5.0
+   Compiling httparse v1.10.1
+   Compiling ring v0.17.14
+   Compiling tracing-core v0.1.36
+   Compiling phf_codegen v0.11.3
+   Compiling block-buffer v0.10.4
+   Compiling alloc-no-stdlib v2.0.4
+   Compiling zerofrom-derive v0.1.7
+   Compiling yoke-derive v0.8.2
+   Compiling zerovec-derive v0.11.3
+   Compiling displaydoc v0.2.6
+   Compiling serde_derive v1.0.228
+   Compiling tokio-macros v2.7.0
+   Compiling futures-macro v0.3.32
+   Compiling thiserror-impl v2.0.18
+   Compiling deno_error_macro v0.6.1
+   Compiling sys_traits_macros v0.1.0
+   Compiling zerofrom v0.1.8
+   Compiling yoke v0.8.3
+   Compiling tokio v1.52.3
+   Compiling zerotrie v0.2.4
+   Compiling zerovec v0.11.6
+   Compiling futures-util v0.3.32
+   Compiling tinystr v0.8.3
+   Compiling potential_utf v0.1.5
+   Compiling icu_locale_core v2.2.0
+   Compiling icu_collections v2.2.0
+   Compiling cpufeatures v0.2.17
+   Compiling zeroize v1.9.0
+   Compiling anyhow v1.0.102
+   Compiling icu_provider v2.2.0
+   Compiling getrandom v0.4.2
+   Compiling icu_properties v2.2.0
+   Compiling icu_normalizer v2.2.0
+   Compiling idna_adapter v1.2.2
+   Compiling idna v1.1.0
+   Compiling rustls-pki-types v1.14.1
+   Compiling url v2.5.8
+   Compiling v8 v137.3.0
+   Compiling alloc-stdlib v0.2.2
+   Compiling tracing-attributes v0.1.31
+   Compiling digest v0.10.7
+   Compiling http-body v1.0.1
+   Compiling new_debug_unreachable v1.0.6
+   Compiling simd-adler32 v0.3.9
+   Compiling untrusted v0.9.0
+   Compiling proc-macro-rules-macros v0.4.0
+   Compiling tracing v0.1.44
+   Compiling deno_error v0.6.1
+   Compiling brotli-decompressor v5.0.1
+   Compiling phf_macros v0.11.3
+   Compiling strum_macros v0.27.2
+   Compiling capacity_builder_macros v0.3.0
+   Compiling pin-project-internal v1.1.13
+   Compiling block-padding v0.3.3
+   Compiling tap v1.0.1
+   Compiling rustls v0.23.40
+   Compiling tower-service v0.3.3
+   Compiling try-lock v0.2.5
+   Compiling utf-8 v0.7.6
+   Compiling wyz v0.5.1
+   Compiling want v0.3.1
+   Compiling brotli v8.0.3
+   Compiling inout v0.1.4
+   Compiling phf v0.11.3
+   Compiling num-integer v0.1.46
+   Compiling flate2 v1.1.9
+   Compiling rustls-webpki v0.103.13
+   Compiling proc-macro-rules v0.4.0
+   Compiling string_cache_codegen v0.5.4
+   Compiling stringcase v0.4.0
+   Compiling funty v2.0.0
+   Compiling compression-core v0.4.32
+   Compiling uuid v1.23.3
+   Compiling atomic-waker v1.1.2
+   Compiling mac v0.1.1
+   Compiling outref v0.5.2
+   Compiling precomputed-hash v0.1.1
+   Compiling vsimd v0.8.0
+   Compiling futf v0.1.5
+   Compiling hyper v1.10.1
+   Compiling debugid v0.8.0
+   Compiling bitvec v1.0.1
+   Compiling markup5ever v0.14.1
+   Compiling strum v0.27.2
+   Compiling base64-simd v0.8.0
+   Compiling deno_ops v0.226.0
+   Compiling futures-executor v0.3.32
+   Compiling compression-codecs v0.4.38
+   Compiling num-bigint v0.4.6
+   Compiling cipher v0.4.4
+   Compiling sys_traits v0.1.28
+   Compiling sync_wrapper v1.0.2
+   Compiling tower-layer v0.3.3
+   Compiling ipnet v2.12.0
+   Compiling dtoa v1.0.11
+   Compiling unicode-id-start v1.4.0
+   Compiling data-encoding v2.11.0
+   Compiling bit-vec v0.8.0
+   Compiling base64 v0.22.1
+   Compiling if_chain v1.0.3
+   Compiling zerocopy v0.8.52
+   Compiling rustix v1.1.4
+   Compiling getrandom v0.3.4
+   Compiling serde_v8 v0.259.0
+   Compiling hyper-util v0.1.20
+   Compiling bit-set v0.8.0
+   Compiling sourcemap v9.3.2
+   Compiling dtoa-short v0.3.5
+   Compiling tower v0.5.3
+   Compiling deno_path_util v0.4.0
+   Compiling tokio-rustls v0.26.4
+   Compiling async-compression v0.4.42
+   Compiling futures v0.3.32
+   Compiling tendril v0.4.3
+   Compiling string_cache v0.8.9
+   Compiling deno_unsync v0.4.4
+   Compiling pin-project v1.1.13
+   Compiling capacity_builder v0.5.0
+   Compiling wasm_dep_analyzer v0.3.0
+   Compiling tokio-util v0.7.18
+   Compiling http-body-util v0.1.3
+   Compiling webpki-roots v1.0.7
+   Compiling bincode v1.3.3
+   Compiling cssparser-macros v0.6.1
+   Compiling selectors v0.26.0
+   Compiling universal-hash v0.5.1
+   Compiling byteorder v1.5.0
+   Compiling deno_core_icudata v0.74.0
+   Compiling ryu v1.0.23
+   Compiling cooked-waker v5.0.0
+   Compiling opaque-debug v0.3.1
+   Compiling az v1.3.0
+   Compiling static_assertions v1.1.0
+   Compiling linux-raw-sys v0.12.1
+   Compiling serde_urlencoded v0.7.1
+   Compiling polyval v0.6.2
+   Compiling fxhash v0.2.1
+   Compiling cssparser v0.34.0
+   Compiling tower-http v0.6.11
+   Compiling hyper-rustls v0.27.9
+   Compiling hmac v0.12.1
+   Compiling sha1 v0.10.6
+   Compiling deno_core v0.350.0
+   Compiling derive_more v0.99.20
+   Compiling match_token v0.1.0
+   Compiling servo_arc v0.4.3
+   Compiling fastrand v2.4.1
+   Compiling tempfile v3.27.0
+   Compiling html5ever v0.29.1
+   Compiling rand_core v0.9.5
+   Compiling reqwest v0.12.28
+   Compiling ghash v0.5.1
+   Compiling ctr v0.9.2
+   Compiling aes v0.8.4
+   Compiling async-trait v0.1.89
+   Compiling aead v0.5.2
+   Compiling encoding_rs v0.8.35
+   Compiling aes-gcm v0.10.3
+   Compiling ppv-lite86 v0.2.21
+   Compiling obscura-dom v0.1.0 (/root/obscura2/merge471/crates/obscura-dom)
+   Compiling hkdf v0.12.4
+   Compiling pbkdf2 v0.12.2
+   Compiling cbc v0.1.2
+   Compiling sha2 v0.10.9
+   Compiling rand_chacha v0.9.0
+   Compiling utf8parse v0.2.2
+   Compiling anstyle-parse v1.0.0
+   Compiling rand v0.9.4
+   Compiling obscura-net v0.1.0 (/root/obscura2/merge471/crates/obscura-net)
+   Compiling colorchoice v1.0.5
+   Compiling anstyle v1.0.14
+   Compiling anstyle-query v1.1.5
+   Compiling is_terminal_polyfill v1.70.2
+   Compiling anstream v1.0.0
+   Compiling tungstenite v0.26.2
+   Compiling strsim v0.11.1
+   Compiling lazy_static v1.5.0
+   Compiling clap_lex v1.1.0
+   Compiling obscura-js v0.1.0 (/root/obscura2/merge471/crates/obscura-js)
+   Compiling clap_builder v4.6.0
+   Compiling sharded-slab v0.1.7
+   Compiling tokio-tungstenite v0.26.2
+   Compiling tracing-log v0.2.0
+   Compiling clap_derive v4.6.1
+   Compiling thread_local v1.1.9
+   Compiling nu-ansi-term v0.50.3
+   Compiling obscura-cli v0.1.0 (/root/obscura2/merge471/crates/obscura-cli)
+   Compiling matchers v0.2.0
+   Compiling tracing-subscriber v0.3.23
+   Compiling clap v4.6.1
+   Compiling obscura-browser v0.1.0 (/root/obscura2/merge471/crates/obscura-browser)
+   Compiling obscura-cdp v0.1.0 (/root/obscura2/merge471/crates/obscura-cdp)
+   Compiling obscura-mcp v0.1.0 (/root/obscura2/merge471/crates/obscura-mcp)
+    Finished `release` profile [optimized] target(s) in 1m 52s

crates/obscura-browser/src/page.rs

@@ -744,6 +744,11 @@ impl Page {
         }
 
         if let Some(js) = &mut self.js {
+            let dynamic_settle_ms = std::env::var("OBSCURA_DYNAMIC_SCRIPT_SETTLE_MS")
+                .ok()
+                .and_then(|s| s.parse::<u64>().ok())
+                .unwrap_or(3_000)
+                .max(500);
             // Bound the post-script settle loop by wall clock, not just by the
             // 10ms-tick branch. The old code only consulted `deadline` inside
             // the `Err(_)` arm (when the inner tick timed out), so a steady
@@ -753,14 +758,22 @@ impl Page {
             // On busy sites this could keep the V8 lock held for tens of
             // seconds, wedging the entire CDP dispatcher (see triage for
             // issue series around the 40-site compat sweep).
+            // A dynamic external script may still be in flight at 500ms. Keep
+            // pumping only while that queue is pending, up to a separate bounded
+            // budget, so normal pages and unrelated fetches retain the fast path.
             // A single run_event_loop poll that pins the thread inside V8 makes
             // the per-poll tokio timeouts below useless, so guard the whole loop
-            // with a watchdog that fires ~250ms past its 500ms deadline.
-            let settle_wd = js.arm_watchdog(std::time::Duration::from_millis(750));
-            let deadline = tokio::time::Instant::now() + tokio::time::Duration::from_millis(500);
+            // with a watchdog that fires 250ms past the longest deadline.
+            let settle_wd = js.arm_watchdog(std::time::Duration::from_millis(dynamic_settle_ms + 250));
+            let started = tokio::time::Instant::now();
+            let deadline = started + tokio::time::Duration::from_millis(500);
+            let dynamic_deadline = started + tokio::time::Duration::from_millis(dynamic_settle_ms);
             let mut idle_count = 0u32;
             loop {
-                if tokio::time::Instant::now() >= deadline {
+                let now = tokio::time::Instant::now();
+                if now >= deadline
+                    && (now >= dynamic_deadline || !js.has_pending_dynamic_scripts())
+                {
                     break;
                 }
                 let result = tokio::time::timeout(

crates/obscura-cdp/tests/dynamic_script_onload_fires.rs

@@ -0,0 +1,119 @@
+// Regression for issue #474: external scripts inserted after a timer must be
+// fetched and execute before the post-navigation event-loop settle completes.
+
+use obscura_cdp::dispatch::{dispatch, CdpContext};
+use obscura_cdp::types::CdpRequest;
+use serde_json::{json, Value};
+use tokio::io::{AsyncReadExt, AsyncWriteExt};
+use tokio::net::TcpListener;
+
+async fn serve() -> String {
+    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    tokio::spawn(async move {
+        loop {
+            let (mut socket, _) = listener.accept().await.unwrap();
+            tokio::spawn(async move {
+                let mut buf = [0u8; 2048];
+                let read = socket.read(&mut buf).await.unwrap();
+                let request = String::from_utf8_lossy(&buf[..read]);
+                let (content_type, body) = if request.starts_with("GET /direct.js") {
+                    tokio::time::sleep(std::time::Duration::from_millis(600)).await;
+                    ("application/javascript", "window.__directExecuted = true;")
+                } else if request.starts_with("GET /nested.js") {
+                    tokio::time::sleep(std::time::Duration::from_millis(50)).await;
+                    ("application/javascript", "window.__nestedExecuted = true;")
+                } else {
+                    (
+                        "text/html",
+                        r#"<html><body>
+<div id="r">stage1</div>
+<script>
+setTimeout(function () {
+  var direct = document.createElement("script");
+  direct.src = "/direct.js";
+  direct.onload = function () { window.__directLoaded = true; };
+  document.body.appendChild(direct);
+
+  var box = document.createElement("div");
+  var nested = document.createElement("script");
+  nested.src = "/nested.js";
+  nested.onload = function () { window.__nestedLoaded = true; };
+  box.appendChild(nested);
+  document.body.appendChild(box);
+}, 100);
+</script>
+</body></html>"#,
+                    )
+                };
+                let response = format!(
+                    "HTTP/1.1 200 OK\r\nContent-Type: {content_type}\r\nContent-Length: {}\r\nConnection: close\r\n\r\n{body}",
+                    body.len()
+                );
+                let _ = socket.write_all(response.as_bytes()).await;
+            });
+        }
+    });
+    format!("http://{addr}/")
+}
+
+async fn cdp(
+    ctx: &mut CdpContext,
+    id: u64,
+    method: &str,
+    params: Value,
+    session_id: &str,
+) -> Value {
+    let response = dispatch(
+        &CdpRequest {
+            id,
+            method: method.to_string(),
+            params,
+            session_id: Some(session_id.to_string()),
+        },
+        ctx,
+    )
+    .await;
+    assert!(
+        response.error.is_none(),
+        "CDP {method} failed: {:?}",
+        response.error
+    );
+    response.result.unwrap_or_else(|| json!({}))
+}
+
+#[tokio::test(flavor = "current_thread")]
+async fn dynamic_external_scripts_execute_and_fire_load() {
+    std::env::set_var("OBSCURA_ALLOW_PRIVATE_NETWORK", "1");
+    let url = serve().await;
+    let mut ctx = CdpContext::new();
+    let page_id = ctx.create_page();
+    let session_id = "session-1";
+    ctx.sessions.insert(session_id.to_string(), page_id);
+
+    cdp(
+        &mut ctx,
+        1,
+        "Page.navigate",
+        json!({"url": url, "waitUntil": "load"}),
+        session_id,
+    )
+    .await;
+
+    let result = cdp(
+        &mut ctx,
+        2,
+        "Runtime.evaluate",
+        json!({
+            "expression": "JSON.stringify({directExecuted: !!window.__directExecuted, directLoaded: !!window.__directLoaded, nestedExecuted: !!window.__nestedExecuted, nestedLoaded: !!window.__nestedLoaded})",
+            "returnByValue": true,
+        }),
+        session_id,
+    )
+    .await;
+    assert_eq!(
+        result["result"]["value"],
+        r#"{"directExecuted":true,"directLoaded":true,"nestedExecuted":true,"nestedLoaded":true}"#,
+        "dynamic scripts must execute and fire load before navigation settles"
+    );
+}

crates/obscura-cdp/tests/iframe_event_dispatch.rs

@@ -0,0 +1,170 @@
+// Regression tests for iframe event dispatch. `_IframeDocument`'s
+// addEventListener/removeEventListener/dispatchEvent were no-ops, so listeners
+// registered on an iframe document never ran. Separately, iframe load invoked
+// the `onload` property directly instead of dispatching through the element,
+// which skipped any addEventListener('load', ...) listener. Both fail on main,
+// pass after the fix.
+
+use obscura_cdp::dispatch::{dispatch, CdpContext};
+use obscura_cdp::types::CdpRequest;
+use serde_json::{json, Value};
+use tokio::io::{AsyncReadExt, AsyncWriteExt};
+use tokio::net::TcpListener;
+
+// Serve every request (not just the first): Page.navigate consumes one, and the
+// iframe load test needs its own successful response so it exercises the real
+// load path rather than the fetch-failure fallback.
+async fn serve() -> String {
+    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    tokio::spawn(async move {
+        while let Ok((mut socket, _)) = listener.accept().await {
+            tokio::spawn(async move {
+                let mut buf = [0u8; 2048];
+                let _ = socket.read(&mut buf).await;
+                let body = "<html><body><div id=a></div></body></html>";
+                let resp = format!(
+                    "HTTP/1.1 200 OK\r\nContent-Type: text/html\r\nContent-Length: {}\r\nConnection: close\r\n\r\n{body}",
+                    body.len()
+                );
+                let _ = socket.write_all(resp.as_bytes()).await;
+            });
+        }
+    });
+    format!("http://{addr}/")
+}
+
+async fn cdp(
+    ctx: &mut CdpContext,
+    id: u64,
+    method: &str,
+    params: Value,
+    session_id: &str,
+) -> Value {
+    let resp = dispatch(
+        &CdpRequest {
+            id,
+            method: method.to_string(),
+            params,
+            session_id: Some(session_id.to_string()),
+        },
+        ctx,
+    )
+    .await;
+    assert!(resp.error.is_none(), "CDP {method} failed: {:?}", resp.error);
+    resp.result.unwrap_or_else(|| json!({}))
+}
+
+async fn eval(ctx: &mut CdpContext, id: u64, expr: &str, session_id: &str) -> Value {
+    cdp(
+        ctx,
+        id,
+        "Runtime.evaluate",
+        json!({"expression": expr, "returnByValue": true, "awaitPromise": true}),
+        session_id,
+    )
+    .await
+}
+
+async fn setup() -> (CdpContext, String) {
+    std::env::set_var("OBSCURA_ALLOW_PRIVATE_NETWORK", "1");
+    let url = serve().await;
+    let mut ctx = CdpContext::new();
+    let page_id = ctx.create_page();
+    let session_id = "session-1";
+    ctx.sessions.insert(session_id.to_string(), page_id.clone());
+    cdp(
+        &mut ctx,
+        1,
+        "Page.navigate",
+        json!({"url": url, "waitUntil": "load"}),
+        session_id,
+    )
+    .await;
+    (ctx, session_id.to_string())
+}
+
+#[tokio::test(flavor = "current_thread")]
+async fn iframe_document_dispatches_registered_listeners() {
+    let (mut ctx, sid) = setup().await;
+    // Accessing contentDocument on a src-less iframe lazily creates an
+    // about:blank _IframeDocument synchronously, so this exercises the event
+    // target without waiting on an async load.
+    let v = eval(
+        &mut ctx,
+        2,
+        r#"(function () {
+            const iframe = document.createElement('iframe');
+            document.body.appendChild(iframe);
+            const doc = iframe.contentDocument;
+            const out = { hasDoc: !!doc };
+            let calls = 0;
+            const listener = () => { calls++; };
+            // A registered listener runs on dispatch.
+            doc.addEventListener('probe', listener);
+            doc.dispatchEvent(new Event('probe'));
+            out.afterRegister = calls;
+            // Registering the same listener again is a no-op (deduped).
+            doc.addEventListener('probe', listener);
+            doc.addEventListener('probe', listener);
+            doc.dispatchEvent(new Event('probe'));
+            out.afterDuplicate = calls;
+            // After removal the listener no longer runs.
+            doc.removeEventListener('probe', listener);
+            doc.dispatchEvent(new Event('probe'));
+            out.afterRemove = calls;
+            // dispatchEvent reports cancellation via its return value.
+            doc.addEventListener('cancelme', (e) => e.preventDefault());
+            out.cancelReturn = doc.dispatchEvent(new Event('cancelme', { cancelable: true }));
+            out.plainReturn = doc.dispatchEvent(new Event('nolisteners'));
+            return JSON.stringify(out);
+        })()"#,
+        &sid,
+    )
+    .await;
+    let val = serde_json::from_str::<Value>(v["result"]["value"].as_str().unwrap()).unwrap();
+    assert_eq!(val["hasDoc"], true);
+    assert_eq!(val["afterRegister"].as_u64(), Some(1), "listener runs once on dispatch");
+    assert_eq!(val["afterDuplicate"].as_u64(), Some(2), "duplicate registration deduped");
+    assert_eq!(val["afterRemove"].as_u64(), Some(2), "removed listener does not run");
+    assert_eq!(val["cancelReturn"], false, "preventDefault -> dispatchEvent returns false");
+    assert_eq!(val["plainReturn"], true, "no cancellation -> dispatchEvent returns true");
+}
+
+#[tokio::test(flavor = "current_thread")]
+async fn iframe_load_reaches_onload_and_addeventlistener() {
+    let (mut ctx, sid) = setup().await;
+    // Both the onload property and an addEventListener('load', ...) listener
+    // must run exactly once. On main, setting onload made the direct-call path
+    // skip the addEventListener listener entirely.
+    let v = eval(
+        &mut ctx,
+        2,
+        r#"(function () {
+            return new Promise((resolve) => {
+                const iframe = document.createElement('iframe');
+                const events = [];
+                // Resolve from the onload property (which fires on both old and
+                // new code) one microtask later, so the synchronous
+                // addEventListener('load') handler has already run on the fixed
+                // build. On main the listener never fires, so this reports
+                // ["property"] (a clean assertion failure) instead of hanging.
+                iframe.onload = () => {
+                    events.push('property');
+                    Promise.resolve().then(() => resolve(JSON.stringify({ events })));
+                };
+                iframe.addEventListener('load', () => events.push('listener'));
+                document.body.appendChild(iframe);
+                iframe.src = location.href;
+            });
+        })()"#,
+        &sid,
+    )
+    .await;
+    let val = serde_json::from_str::<Value>(v["result"]["value"].as_str().unwrap()).unwrap();
+    assert_eq!(
+        val["events"],
+        json!(["property", "listener"]),
+        "both onload property and addEventListener('load') run exactly once, in order"
+    );
+}

crates/obscura-cdp/tests/scroll_event_on_assignment.rs

@@ -0,0 +1,165 @@
+//! Assigning `scrollTop` / `scrollLeft` must fire a `scroll` event.
+//!
+//! `scrollTo`/`scrollBy` already dispatched one; direct assignment did not, so
+//! the very common lazy-load idiom
+//!
+//! ```js
+//! el.addEventListener('scroll', loadMore);
+//! el.scrollTop = el.scrollHeight;      // never reached loadMore
+//! ```
+//!
+//! silently did nothing and scroll-driven feeds stalled after their first batch.
+//!
+//! Checked against a real Chrome over CDP with this same probe: assigning
+//! `scrollTop` fires exactly one `scroll` event there. These tests assert the
+//! same observable behaviour, not the synthetic geometry around it — the offset
+//! is intentionally not clamped, since without layout any maximum is a guess.
+
+use obscura_cdp::dispatch::{dispatch, CdpContext};
+use obscura_cdp::types::CdpRequest;
+use serde_json::{json, Value};
+use tokio::io::{AsyncReadExt, AsyncWriteExt};
+use tokio::net::TcpListener;
+
+async fn serve_once() -> String {
+    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    tokio::spawn(async move {
+        let (mut socket, _) = listener.accept().await.unwrap();
+        tokio::spawn(async move {
+            let mut buf = [0u8; 2048];
+            let _ = socket.read(&mut buf).await.unwrap();
+            let body = "<html><body><div id=\"box\"><p>a</p><p>b</p></div></body></html>";
+            let resp = format!(
+                "HTTP/1.1 200 OK\r\nContent-Type: text/html\r\nContent-Length: {}\r\nConnection: close\r\n\r\n{body}",
+                body.len()
+            );
+            let _ = socket.write_all(resp.as_bytes()).await;
+        });
+    });
+    format!("http://{addr}/")
+}
+
+async fn cdp(ctx: &mut CdpContext, id: u64, method: &str, params: Value, session_id: &str) -> Value {
+    let resp = dispatch(
+        &CdpRequest {
+            id,
+            method: method.to_string(),
+            params,
+            session_id: Some(session_id.to_string()),
+        },
+        ctx,
+    )
+    .await;
+    assert!(resp.error.is_none(), "CDP {method} failed: {:?}", resp.error);
+    resp.result.unwrap_or_else(|| json!({}))
+}
+
+/// `awaitPromise` matters here: the scroll event is dispatched from a
+/// `setTimeout(..., 0)`, so the probe has to yield before reading the counter.
+async fn eval(ctx: &mut CdpContext, id: u64, expr: &str, session_id: &str) -> Value {
+    cdp(
+        ctx,
+        id,
+        "Runtime.evaluate",
+        json!({"expression": expr, "returnByValue": true, "awaitPromise": true}),
+        session_id,
+    )
+    .await
+}
+
+async fn setup() -> (CdpContext, String) {
+    std::env::set_var("OBSCURA_ALLOW_PRIVATE_NETWORK", "1");
+    let url = serve_once().await;
+    let mut ctx = CdpContext::new();
+    let page_id = ctx.create_page();
+    let session_id = "session-1";
+    ctx.sessions.insert(session_id.to_string(), page_id.clone());
+    cdp(&mut ctx, 1, "Page.navigate", json!({"url": url, "waitUntil": "load"}), session_id).await;
+    (ctx, session_id.to_string())
+}
+
+async fn probe(ctx: &mut CdpContext, sid: &str, body: &str) -> Value {
+    let expr = format!(
+        r#"(() => {{
+            const el = document.getElementById('box');
+            let fired = 0;
+            el.addEventListener('scroll', () => {{ fired++; }});
+            {body}
+            return new Promise(r => setTimeout(() => r(JSON.stringify({{
+                fired, top: el.scrollTop, left: el.scrollLeft,
+            }})), 0));
+        }})()"#
+    );
+    let v = eval(ctx, 2, &expr, sid).await;
+    serde_json::from_str::<Value>(v["result"]["value"].as_str().unwrap()).unwrap()
+}
+
+#[tokio::test(flavor = "current_thread")]
+async fn assigning_scroll_top_fires_one_scroll_event() {
+    let (mut ctx, sid) = setup().await;
+    let r = probe(&mut ctx, &sid, "el.scrollTop = 100;").await;
+    assert_eq!(r["fired"], 1, "scrollTop assignment must fire exactly one scroll event");
+    assert_eq!(r["top"], 100, "scrollTop must round-trip the assigned value");
+}
+
+#[tokio::test(flavor = "current_thread")]
+async fn assigning_scroll_left_fires_one_scroll_event() {
+    let (mut ctx, sid) = setup().await;
+    let r = probe(&mut ctx, &sid, "el.scrollLeft = 40;").await;
+    assert_eq!(r["fired"], 1, "scrollLeft assignment must fire exactly one scroll event");
+    assert_eq!(r["left"], 40, "scrollLeft must round-trip the assigned value");
+}
+
+/// Re-assigning the same offset is not a scroll, so it must stay silent —
+/// otherwise a loader that writes `scrollTop` on every frame re-enters forever.
+#[tokio::test(flavor = "current_thread")]
+async fn reassigning_the_same_offset_is_silent() {
+    let (mut ctx, sid) = setup().await;
+    let r = probe(&mut ctx, &sid, "el.scrollTop = 100; el.scrollTop = 100;").await;
+    assert_eq!(r["fired"], 1, "only the offset change fires; the repeat must not");
+}
+
+/// `scrollTo` moves both axes, and a real browser reports one scroll per
+/// movement rather than one per axis. The setters suppress their own event
+/// inside `scrollTo`/`scrollBy` so the operation stays a single event.
+#[tokio::test(flavor = "current_thread")]
+async fn scroll_to_coalesces_both_axes_into_one_event() {
+    let (mut ctx, sid) = setup().await;
+    let r = probe(&mut ctx, &sid, "el.scrollTo(30, 60);").await;
+    assert_eq!(r["fired"], 1, "scrollTo must fire one event, not one per axis");
+    assert_eq!(r["top"], 60);
+    assert_eq!(r["left"], 30);
+}
+
+/// The lazy-load idiom the fix exists for: a listener that appends more rows
+/// when the feed is scrolled. Before the fix this counted 0 and feeds froze.
+#[tokio::test(flavor = "current_thread")]
+async fn scroll_driven_lazy_loader_advances() {
+    let (mut ctx, sid) = setup().await;
+    let v = eval(
+        &mut ctx,
+        2,
+        r#"(() => {
+            const el = document.getElementById('box');
+            let batches = 0;
+            el.addEventListener('scroll', () => {
+                if (batches >= 3) return;
+                batches++;
+                for (let i = 0; i < 5; i++) el.appendChild(document.createElement('p'));
+            });
+            el.scrollTop = 500;
+            return new Promise(r => setTimeout(() => {
+                el.scrollTop = 1000;
+                setTimeout(() => r(JSON.stringify({
+                    batches, rows: el.querySelectorAll('p').length,
+                })), 0);
+            }, 0));
+        })()"#,
+        &sid,
+    )
+    .await;
+    let r = serde_json::from_str::<Value>(v["result"]["value"].as_str().unwrap()).unwrap();
+    assert_eq!(r["batches"], 2, "each forward scroll must reach the loader");
+    assert_eq!(r["rows"], 12, "2 starting rows + 2 batches of 5");
+}

crates/obscura-dom/src/serialize.rs

@@ -21,10 +21,21 @@ impl DomTree {
 
     pub fn inner_html(&self, node_id: NodeId) -> String {
         let mut buf = String::new();
-        self.serialize_worklist(self.child_work(node_id), &mut buf);
+        self.serialize_worklist(self.child_work(self.content_source(node_id)), &mut buf);
         buf
     }
 
+    // The node whose children represent `node_id`'s markup. For a <template>
+    // that is its contents document, since the parser puts template children
+    // there rather than under the element (issue #463); for everything else it
+    // is the node itself.
+    fn content_source(&self, node_id: NodeId) -> NodeId {
+        match self.get_node(node_id).map(|n| n.data) {
+            Some(NodeData::Element { template_contents: Some(contents), .. }) => contents,
+            _ => node_id,
+        }
+    }
+
     // Children as work items in document order (top of the LIFO stack first).
     fn child_work(&self, node_id: NodeId) -> Vec<SerializeWork> {
         self.children(node_id)
@@ -101,7 +112,9 @@ impl DomTree {
                         if include_self {
                             stack.push(SerializeWork::CloseTag(tag.to_string()));
                         }
-                        for w in self.child_work(node_id) {
+                        // A <template> serializes its contents document, not its
+                        // own (always empty) children — issue #463.
+                        for w in self.child_work(self.content_source(node_id)) {
                             stack.push(w);
                         }
                     }

crates/obscura-dom/src/tree.rs

@@ -579,6 +579,41 @@ impl DomTree {
         Self::climb_to_next_sibling(&inner, root, current)
     }
 
+    /// Returns the node before `current` in document order, without leaving the
+    /// subtree rooted at `root`. `root` has no predecessor within its own
+    /// subtree, but it is itself reachable as one — a NodeIterator can return
+    /// its root, unlike a TreeWalker.
+    ///
+    /// A NodeIterator applies no subtree pruning (DOM 6.2: FILTER_REJECT
+    /// behaves as FILTER_SKIP), so unlike the TreeWalker's backward walk the
+    /// whole step fits here instead of being interleaved with filter calls.
+    pub fn prev_in_subtree(&self, root: NodeId, current: NodeId) -> Option<NodeId> {
+        let inner = self.inner.borrow();
+        if current == root {
+            return None;
+        }
+        let current_node = inner.nodes.get(current.index())?.as_ref()?;
+
+        let Some(prev) = current_node.prev_sibling else {
+            // No previous sibling: the parent immediately precedes `current`.
+            return current_node.parent;
+        };
+
+        // Otherwise it is the previous sibling's deepest last descendant.
+        let mut node_id = prev;
+        for _ in 0..=inner.nodes.len() {
+            let node = inner.nodes.get(node_id.index())?.as_ref()?;
+            match node.last_child {
+                Some(child) => node_id = child,
+                None => return Some(node_id),
+            }
+        }
+
+        // Same defense in depth as the forward walk: a malformed tree must not
+        // spin here.
+        None
+    }
+
     /// Follow `current`'s next sibling, climbing ancestors until one has a next
     /// sibling — without stepping outside `root`.
     fn climb_to_next_sibling(
@@ -603,6 +638,41 @@ impl DomTree {
         None
     }
 
+    /// The node holding a `<template>` element's contents.
+    ///
+    /// The parser puts template children in a separate contents document rather
+    /// than under the element (HTML spec), so this is the only way to reach
+    /// them. Templates built with `createElement` have no contents node yet, so
+    /// one is allocated on demand — `.content` must be usable either way.
+    ///
+    /// Returns `None` for a non-element node.
+    pub fn template_contents(&self, node_id: NodeId) -> Option<NodeId> {
+        {
+            let inner = self.inner.borrow();
+            let node = inner.nodes.get(node_id.index())?.as_ref()?;
+            match &node.data {
+                NodeData::Element { template_contents, .. } => {
+                    if let Some(existing) = *template_contents {
+                        return Some(existing);
+                    }
+                }
+                _ => return None,
+            }
+        }
+
+        // Borrow released above: `new_node` takes its own mutable borrow.
+        // Matches what the tree sink allocates for a parsed template.
+        let contents = self.new_node(NodeData::Document);
+        let mut inner = self.inner.borrow_mut();
+        if let Some(Some(node)) = inner.nodes.get_mut(node_id.index()) {
+            if let NodeData::Element { template_contents, .. } = &mut node.data {
+                *template_contents = Some(contents);
+                return Some(contents);
+            }
+        }
+        None
+    }
+
     pub fn ancestors(&self, node_id: NodeId) -> Vec<NodeId> {
         let inner = self.inner.borrow();
         let mut result = Vec::new();
@@ -727,6 +797,38 @@ impl DomTree {
             let new_id = self.new_node(node_data);
             self.append_child(dest_parent, new_id);
 
+            // A <template>'s children hang off a separate contents document, so
+            // the child walk below never reaches them. Worse, the cloned data
+            // carries the *source* tree's contents NodeId, which here indexes
+            // whatever unrelated node occupies that slot. Allocate a real
+            // contents node and queue the source contents' children into it, so
+            // the reference is remapped rather than left dangling (issue #463).
+            let src_contents = {
+                let inner = self.inner.borrow();
+                match inner.nodes.get(new_id.index()).and_then(|n| n.as_ref()) {
+                    Some(node) => match &node.data {
+                        NodeData::Element { template_contents, .. } => *template_contents,
+                        _ => None,
+                    },
+                    None => None,
+                }
+            };
+            if let Some(src_contents) = src_contents {
+                let dest_contents = self.new_node(NodeData::Document);
+                {
+                    let mut inner = self.inner.borrow_mut();
+                    if let Some(Some(node)) = inner.nodes.get_mut(new_id.index()) {
+                        if let NodeData::Element { template_contents, .. } = &mut node.data {
+                            *template_contents = Some(dest_contents);
+                        }
+                    }
+                }
+                // Onto the same stack, so nested templates stay iterative.
+                for child_id in source.children(src_contents).into_iter().rev() {
+                    stack.push((dest_contents, child_id));
+                }
+            }
+
             for child_id in source.children(src_id).into_iter().rev() {
                 stack.push((new_id, child_id));
             }
@@ -1071,6 +1173,29 @@ mod tests {
         assert_eq!(tree.next_after_subtree(root, root), None);
     }
 
+    #[test]
+    fn test_prev_in_subtree_reverses_document_order() {
+        // root > [first > nested, second]; document order is root, first,
+        // nested, second, so the reverse walk must retrace it exactly.
+        let tree = DomTree::new();
+        let root = tree.new_node(NodeData::Text { contents: "root".into() });
+        let first = tree.new_node(NodeData::Text { contents: "first".into() });
+        let nested = tree.new_node(NodeData::Text { contents: "nested".into() });
+        let second = tree.new_node(NodeData::Text { contents: "second".into() });
+        tree.append_child(tree.document(), root);
+        tree.append_child(root, first);
+        tree.append_child(first, nested);
+        tree.append_child(root, second);
+
+        // Previous sibling's deepest last descendant, not the sibling itself.
+        assert_eq!(tree.prev_in_subtree(root, second), Some(nested));
+        assert_eq!(tree.prev_in_subtree(root, nested), Some(first));
+        // No previous sibling: the parent precedes it, and root is returnable.
+        assert_eq!(tree.prev_in_subtree(root, first), Some(root));
+        // Root has no predecessor inside its own subtree.
+        assert_eq!(tree.prev_in_subtree(root, root), None);
+    }
+
     // Builds a chain of `depth` nested <div> elements under the document and
     // returns (outermost_div, innermost_div). Depth this large overflows a
     // recursive tree walk and aborts the process, so it guards the iterative

crates/obscura-js/js/bootstrap.js

@@ -1216,6 +1216,8 @@ class Element extends Node {
     if (this.localName === "svg") return "http://www.w3.org/2000/svg";
     return "http://www.w3.org/1999/xhtml";
   }
+  // `inner_html` resolves a <template> to its contents document on the Rust
+  // side (issue #463), so this needs no template special case.
   get innerHTML() { return _domParse("inner_html", this._nid) ?? ""; }
   set innerHTML(v) {
     if (this.localName === 'template') {
@@ -1254,6 +1256,18 @@ class Element extends Node {
     // infinite retry loop (issue #210).
     const tag = this.localName;
     if (tag === 'template') {
+      // Back the fragment with the node's real template contents (issue #463).
+      // The parser stores template children in a separate contents document
+      // instead of under the element, so without this the getter handed back a
+      // fabricated empty fragment and the parsed markup was unreachable.
+      // `template_contents` allocates one on demand for created templates.
+      const nid = +_dom("template_contents", this._nid);
+      if (nid >= 0) {
+        // Cache by node id so `.content` keeps a stable identity across reads —
+        // frameworks stash the fragment and compare it later.
+        if (!_cache.has(nid)) _cache.set(nid, new DocumentFragment(nid));
+        return _cache.get(nid);
+      }
       if (!this._templateContent) this._templateContent = document.createDocumentFragment();
       return this._templateContent;
     }
@@ -1929,17 +1943,15 @@ class Element extends Node {
         el._iframeWin = new _IframeWindow(el._iframeDoc, fullUrl);
       }
 
-      if (typeof el.onload === 'function') {
-        try { el.onload(); } catch(e) {}
-      } else {
-        var onloadAttr = el.getAttribute('onload');
-        if (onloadAttr) try { (0, eval)(onloadAttr); } catch(e) {}
-      }
+      // Dispatch through the element so the onload property/attribute and any
+      // addEventListener('load', ...) listeners all run. Calling el.onload()
+      // directly bypasses listeners registered via addEventListener.
+      el.dispatchEvent(new Event('load'));
     }).catch(() => {
       el._iframeDoc = new _IframeDocument('<!DOCTYPE html><html><head></head><body></body></html>', fullUrl, el);
       el._iframeWin = new _IframeWindow(el._iframeDoc, fullUrl);
 
-      if (typeof el.onload === 'function') try { el.onload(); } catch(e) {}
+      el.dispatchEvent(new Event('load'));
     });
   }
   get contentDocument() {
@@ -2110,13 +2122,29 @@ class Element extends Node {
     const t = this.tagName;
     return t === 'HTML' || t === 'BODY';
   }
-  // No layout engine, so there is no real overflow to scroll. We still track a
-  // scroll offset so scrollTop/scrollLeft round-trip and the scroll methods
-  // below can report a position, which is what infinite-scroll code reads back.
+  // No layout engine, so there is no real overflow to scroll and the offset is
+  // deliberately NOT clamped: without real geometry any synthetic max is a
+  // guess, and a max derived from a stub scroll box pins scrollTop at 0, which
+  // deadlocks scroll-driven lazy loaders (no scroll -> no content -> no scroll).
+  // We track the offset so scrollTop/scrollLeft round-trip, and fire a scroll
+  // event on direct assignment -- lazy loaders that set `el.scrollTop = N` rely
+  // on that event, and scrollTo/scrollBy below would otherwise be its only source.
   get scrollTop() { return this._scrollTop || 0; }
-  set scrollTop(v) { v = +v; this._scrollTop = Number.isFinite(v) && v > 0 ? v : 0; }
+  set scrollTop(v) {
+    v = +v;
+    const nv = Number.isFinite(v) && v > 0 ? v : 0;
+    const changed = nv !== (this._scrollTop || 0);
+    this._scrollTop = nv;
+    if (changed && !this._scrollSuppress) this._fireScroll();
+  }
   get scrollLeft() { return this._scrollLeft || 0; }
-  set scrollLeft(v) { v = +v; this._scrollLeft = Number.isFinite(v) && v > 0 ? v : 0; }
+  set scrollLeft(v) {
+    v = +v;
+    const nv = Number.isFinite(v) && v > 0 ? v : 0;
+    const changed = nv !== (this._scrollLeft || 0);
+    this._scrollLeft = nv;
+    if (changed && !this._scrollSuppress) this._fireScroll();
+  }
   getBoundingClientRect() {
     globalThis.__obscura_click_target = this;
     // documentElement and body span the full viewport. Without this every
@@ -2175,15 +2203,17 @@ class Element extends Node {
   set ariaSelected(v) { if (v == null) this.removeAttribute('aria-selected'); else this.setAttribute('aria-selected', String(v)); }
   scrollIntoView() { globalThis.__obscura_click_target = this; }
   // scrollTo/scrollBy/scroll accept either (x, y) or a ScrollToOptions object.
-  // Without layout the offset cannot be clamped to a real max, but updating it
-  // and firing a scroll event lets scroll-driven lazy loaders advance instead
-  // of throwing "scrollBy is not a function".
+  // The setters fire a scroll event of their own, so suppress the per-axis ones
+  // here and emit a single event for the whole movement, the way a real browser
+  // coalesces one scroll per scroll operation rather than one per axis.
   scrollTo(x, y) {
     let left, top;
     if (x !== null && typeof x === 'object') { left = x.left; top = x.top; }
     else { left = x; top = y; }
+    this._scrollSuppress = true;
     if (left !== undefined) this.scrollLeft = +left || 0;
     if (top !== undefined) this.scrollTop = +top || 0;
+    this._scrollSuppress = false;
     this._fireScroll();
   }
   scroll(x, y) { this.scrollTo(x, y); }
@@ -2191,8 +2221,10 @@ class Element extends Node {
     let dl, dt;
     if (x !== null && typeof x === 'object') { dl = x.left; dt = x.top; }
     else { dl = x; dt = y; }
+    this._scrollSuppress = true;
     this.scrollLeft = (this.scrollLeft || 0) + (+dl || 0);
     this.scrollTop = (this.scrollTop || 0) + (+dt || 0);
+    this._scrollSuppress = false;
     this._fireScroll();
   }
   _fireScroll() {
@@ -2637,11 +2669,8 @@ class Document extends Node {
           const verdict = this._filter(node);
           if (verdict === 1) { this.currentNode = node; return node; }
           // FILTER_REJECT skips the node AND its subtree; FILTER_SKIP (and any
-          // other non-accept value) skips only the node. NodeIterator has no
-          // pruning at all, so `_rejectIsSkip` keeps it on the plain step.
-          const step = (verdict === 2 && !this._rejectIsSkip)
-            ? "next_after_subtree"
-            : "next_in_subtree";
+          // other non-accept value) skips only the node.
+          const step = verdict === 2 ? "next_after_subtree" : "next_in_subtree";
           node = _wrap(+_dom(step, this.root._nid, node._nid));
         }
         return null;
@@ -2734,26 +2763,86 @@ class Document extends Node {
       lastChild() { return this._traverseChildren('lastChild', 'previousSibling'); },
       nextSibling() { return this._traverseSiblings('firstChild', 'nextSibling'); },
       previousSibling() { return this._traverseSiblings('lastChild', 'previousSibling'); },
+      // DOM 6.1 "parentNode" (issue #475). The old version looked only at the
+      // immediate parent, so it couldn't climb past a skipped ancestor; it also
+      // excluded `root` as a result yet stepped to root's own parent when
+      // currentNode was root, returning a node OUTSIDE the walker's subtree.
+      // The loop's `node !== this.root` guard is what keeps the walk inside
+      // root while still allowing root itself to be returned.
       parentNode() {
-        let parent = this.currentNode.parentNode;
-        if (parent && parent !== this.root && this._accept(parent)) {
-          this.currentNode = parent;
-          return parent;
+        let node = this.currentNode;
+        while (node && node !== this.root) {
+          node = node.parentNode;
+          if (node && this._accept(node)) { this.currentNode = node; return node; }
         }
         return null;
       },
     };
     return walker;
   }
+  // A real NodeIterator (DOM 6.2), not a TreeWalker in disguise (issue #467).
+  // The two differ in more than naming: an iterator's pointer starts *before*
+  // its root, so the first nextNode() returns the root itself, and it exposes
+  // referenceNode/pointerBeforeReferenceNode/detach rather than a TreeWalker's
+  // currentNode and child/sibling movers.
   createNodeIterator(root, whatToShow, filter) {
-    // Shares the TreeWalker implementation, but DOM 6.2 gives NodeIterator no
-    // subtree pruning: FILTER_REJECT behaves exactly as FILTER_SKIP there.
-    const iterator = this.createTreeWalker(root, whatToShow, filter);
-    iterator._rejectIsSkip = true;
-    return iterator;
+    // whatToShow is unsigned long; default SHOW_ALL only when the arg is
+    // omitted. An explicit 0 (show nothing) must stay 0, not become SHOW_ALL.
+    whatToShow = (whatToShow === undefined) ? 0xFFFFFFFF : (whatToShow >>> 0);
+    return {
+      root: root,
+      referenceNode: root,
+      pointerBeforeReferenceNode: true,
+      whatToShow: whatToShow,
+      filter: filter || null,
+      // NodeIterator prunes nothing: FILTER_REJECT behaves as FILTER_SKIP, so
+      // unlike the TreeWalker only "accepted or not" matters here.
+      _accept(node) {
+        if (!((whatToShow >> (node.nodeType - 1)) & 1)) return false;
+        if (this.filter) {
+          if (typeof this.filter === 'function') return this.filter(node) === 1;
+          if (this.filter.acceptNode) return this.filter.acceptNode(node) === 1;
+        }
+        return true;
+      },
+      // DOM 6.2 "traverse". The pointer sits either before or after
+      // referenceNode, which is why reversing direction re-yields the current
+      // node instead of stepping over it.
+      _traverse(forward) {
+        let node = this.referenceNode;
+        let before = this.pointerBeforeReferenceNode;
+        for (;;) {
+          if (forward === before) {
+            // Consume the pointer's side without moving: it flips to the other
+            // side of the node it already references.
+            before = !before;
+          } else {
+            const step = forward ? "next_in_subtree" : "prev_in_subtree";
+            const next = _wrap(+_dom(step, this.root._nid, node._nid));
+            // A failed traversal leaves referenceNode and the pointer
+            // untouched, so the iterator can be resumed in either direction.
+            if (!next) return null;
+            node = next;
+          }
+          if (this._accept(node)) break;
+        }
+        this.referenceNode = node;
+        this.pointerBeforeReferenceNode = before;
+        return node;
+      },
+      nextNode() { return this._traverse(true); },
+      previousNode() { return this._traverse(false); },
+      // Legacy no-op since DOM4, but older library code still calls it and
+      // used to hit "detach is not a function".
+      detach() {},
+    };
   }
   getSelection() { return this.defaultView ? _selectionFor(this) : null; }
   get activeElement() { return globalThis.__obscura_focused || this.body; }
+  // The element that scrolls the viewport, and where the page offset lives
+  // (issue #468). Standards mode, so documentElement — quirks mode would be
+  // body, but we never parse in quirks mode.
+  get scrollingElement() { return this.documentElement; }
   get implementation() {
     const ownerDoc = this;
     return {
@@ -6152,9 +6241,33 @@ class _IframeDocument {
   get implementation() { return document.implementation; }
   get styleSheets() { return []; }
 
-  addEventListener() {}
-  removeEventListener() {}
-  dispatchEvent() { return true; }
+  addEventListener(type, listener) {
+    if (typeof listener !== 'function') return;
+    if (!this._listeners) this._listeners = Object.create(null);
+    const list = this._listeners[type] || (this._listeners[type] = []);
+    if (!list.includes(listener)) list.push(listener);
+  }
+  removeEventListener(type, listener) {
+    const list = this._listeners && this._listeners[type];
+    if (!list) return;
+    const index = list.indexOf(listener);
+    if (index !== -1) list.splice(index, 1);
+  }
+  dispatchEvent(event) {
+    const type = event && event.type;
+    if (!type) return true;
+    const list = this._listeners && this._listeners[type];
+    if (list) {
+      for (const listener of list.slice()) {
+        try { listener.call(this, event); } catch (error) { console.error(error); }
+      }
+    }
+    const handler = this['on' + type];
+    if (typeof handler === 'function') {
+      try { handler.call(this, event); } catch (error) { console.error(error); }
+    }
+    return !event.defaultPrevented;
+  }
 
   write(html) {
     if (this._body) this._body.innerHTML += html;
@@ -7152,9 +7265,63 @@ URL.revokeObjectURL = function(url) {
   delete globalThis.__blobStore[url];
 };
 
-globalThis.scrollTo = function(x, y) {};
-globalThis.scrollBy = function(x, y) {};
-globalThis.scroll = function(x, y) {};
+// Window-level scrolling (issue #468). #431 gave elements functional
+// scrollTop/scrollLeft plus scroll methods, but left these three as no-ops, so
+// the dominant infinite-scroll idiom -- window.scrollTo(0, body.scrollHeight),
+// window.scrollBy(0, 500), then a window 'scroll' listener -- did nothing at
+// all: the offset never moved and no event ever fired.
+//
+// The page offset is stored on the scrolling element rather than in separate
+// window state, so window.scrollY and document.scrollingElement.scrollTop are
+// two views of one value, which is what pages assume. As with #431 there is no
+// layout, so the offset still cannot be clamped to a real maximum.
+function _scrollRoot() {
+  const doc = globalThis.document;
+  return (doc && doc.scrollingElement) || null;
+}
+function _windowScroll(x, y, relative) {
+  const root = _scrollRoot();
+  if (!root) return;
+  let left, top;
+  if (x !== null && typeof x === 'object') { left = x.left; top = x.top; }
+  else { left = x; top = y; }
+  if (left !== undefined) {
+    root.scrollLeft = (relative ? (root.scrollLeft || 0) : 0) + (+left || 0);
+  }
+  if (top !== undefined) {
+    root.scrollTop = (relative ? (root.scrollTop || 0) : 0) + (+top || 0);
+  }
+  // Async, matching the element path #431 added. Dispatched at the document
+  // AND the window: a page scroll event reaches both in Chrome, but
+  // Document.dispatchEvent here runs only its own listeners and does not
+  // propagate, so firing once would strand half the listeners.
+  setTimeout(() => {
+    try {
+      const doc = globalThis.document;
+      if (doc) doc.dispatchEvent(new Event('scroll', { bubbles: false }));
+      globalThis.dispatchEvent(new Event('scroll', { bubbles: false }));
+    } catch (e) {}
+  }, 0);
+}
+globalThis.scrollTo = function(x, y) { _windowScroll(x, y, false); };
+globalThis.scrollBy = function(x, y) { _windowScroll(x, y, true); };
+globalThis.scroll = function(x, y) { _windowScroll(x, y, false); };
+_markNative(globalThis.scrollTo);
+_markNative(globalThis.scrollBy);
+_markNative(globalThis.scroll);
+// Read-only accessors, as on a real Window: assigning window.scrollY does not
+// scroll the page. These replace the hard-coded 0 data properties defined
+// earlier, so they must stay after them.
+for (const [name, offset] of [
+  ['scrollX', 'scrollLeft'], ['pageXOffset', 'scrollLeft'],
+  ['scrollY', 'scrollTop'], ['pageYOffset', 'scrollTop'],
+]) {
+  Object.defineProperty(globalThis, name, {
+    configurable: true,
+    enumerable: true,
+    get() { const root = _scrollRoot(); return root ? (root[offset] || 0) : 0; },
+  });
+}
 globalThis.focus = function() {}; _markNative(globalThis.focus);
 globalThis.blur = function() {}; _markNative(globalThis.blur);
 globalThis.print = function() {}; _markNative(globalThis.print);

crates/obscura-js/src/ops.rs

@@ -275,6 +275,15 @@ fn op_dom_inner(state: &OpState, cmd: String, arg1: String, arg2: String) -> Str
                 .map(|id| id.index().to_string())
                 .unwrap_or("-1".into())
         }
+        // Reverse document order within a subtree, for NodeIterator's backward
+        // walk (which prunes nothing, so the whole step fits in the DOM layer).
+        "prev_in_subtree" => {
+            let root = NodeId::new(arg1.parse::<u32>().unwrap_or(0));
+            let current = NodeId::new(arg2.parse::<u32>().unwrap_or(0));
+            dom.prev_in_subtree(root, current)
+                .map(|id| id.index().to_string())
+                .unwrap_or("-1".into())
+        }
         // Step past a whole subtree rather than into it: NodeFilter.FILTER_REJECT
         // prunes the rejected node's descendants, unlike FILTER_SKIP.
         "next_after_subtree" => {
@@ -392,6 +401,15 @@ fn op_dom_inner(state: &OpState, cmd: String, arg1: String, arg2: String) -> Str
             });
             "true".into()
         }
+        // A <template>'s children live in a separate contents document, so this
+        // is the only route to them from JS. Allocates one on demand for
+        // templates built via createElement.
+        "template_contents" => {
+            let nid = arg1.parse::<u32>().unwrap_or(0);
+            dom.template_contents(NodeId::new(nid))
+                .map(|id| id.index().to_string())
+                .unwrap_or("-1".into())
+        }
         "create_document_fragment" => {
             dom.new_node(NodeData::Document).index().to_string()
         }

crates/obscura-js/src/runtime.rs

@@ -859,6 +859,18 @@ impl ObscuraJsRuntime {
             .map_err(|e| format!("Event loop error: {}", e))
     }
 
+    /// Whether the serialized dynamic-script queue is still fetching or
+    /// evaluating a script. The queue variables are global lexicals rather
+    /// than window properties, so page code cannot overwrite this state.
+    pub fn has_pending_dynamic_scripts(&mut self) -> bool {
+        self.evaluate(
+            "typeof __dynScriptBusy !== 'undefined' && (__dynScriptBusy || __dynScriptQueue.length > 0)",
+        )
+        .ok()
+        .and_then(|value| value.as_bool())
+        .unwrap_or(false)
+    }
+
     /// Arm a hard wall-clock backstop on synchronous V8 work. A page stuck in a
     /// synchronous loop or a microtask storm pins the OS thread inside V8, so
     /// `tokio::time::timeout` (which can only cancel at await points) never
@@ -1513,6 +1525,75 @@ mod tests {
 
     /// Issue #461: NodeIterator has no subtree pruning — DOM 6.2 says
     /// FILTER_REJECT behaves as FILTER_SKIP there. The shared walker must not
+    /// Issue #475: parentNode() must never surface a node above `root`. With
+    /// currentNode at root, the old guard stepped to root's own parent and
+    /// returned it — escaping the walker's subtree entirely.
+    #[test]
+    fn tree_walker_parent_node_does_not_escape_above_root() {
+        let mut rt = setup_runtime(r#"<div id="root"><a></a></div>"#);
+        let result = rt
+            .evaluate(
+                r#"
+                const root = document.getElementById('root');
+                const w = document.createTreeWalker(root, NodeFilter.SHOW_ELEMENT);
+                const escaped = w.parentNode();
+                return [escaped, w.currentNode.id];
+                "#,
+            )
+            .unwrap();
+        // No parent within the subtree, and currentNode stays put at root.
+        assert_eq!(result, serde_json::json!([null, "root"]));
+    }
+
+    /// Issue #475: when the accepted ancestor is `root` itself, parentNode()
+    /// returns it and moves currentNode there — the old `parent !== root` guard
+    /// wrongly excluded it.
+    #[test]
+    fn tree_walker_parent_node_can_return_the_root() {
+        let mut rt = setup_runtime(r#"<div id="root"><a></a></div>"#);
+        let result = rt
+            .evaluate(
+                r#"
+                const root = document.getElementById('root');
+                const w = document.createTreeWalker(root, NodeFilter.SHOW_ELEMENT);
+                w.currentNode = root.querySelector('a');
+                const p = w.parentNode();
+                return [p ? p.id : null, w.currentNode === root];
+                "#,
+            )
+            .unwrap();
+        assert_eq!(result, serde_json::json!(["root", true]));
+    }
+
+    /// Issue #475: parentNode() climbs past a skipped ancestor to the first
+    /// accepted one, instead of stopping at the immediate parent.
+    #[test]
+    fn tree_walker_parent_node_climbs_past_skipped_ancestors() {
+        let mut rt = setup_runtime(
+            r#"<div id="root"><main id="m"><section><a></a></section></main></div>"#,
+        );
+        let result = rt
+            .evaluate(
+                r#"
+                const root = document.getElementById('root');
+                const w = document.createTreeWalker(root, NodeFilter.SHOW_ELEMENT, {
+                    acceptNode(n) {
+                        return n.tagName === 'SECTION'
+                            ? NodeFilter.FILTER_SKIP
+                            : NodeFilter.FILTER_ACCEPT;
+                    }
+                });
+                w.currentNode = root.querySelector('a');
+                const p = w.parentNode();
+                return p ? p.id : null;
+                "#,
+            )
+            .unwrap();
+        // <a>'s parent <section> is skipped, so <main> is the first accepted
+        // ancestor — not null, and not the immediate <section>.
+        assert_eq!(result, serde_json::json!("m"));
+    }
+
     /// leak TreeWalker's pruning into it.
     #[test]
     fn node_iterator_treats_filter_reject_as_skip() {
@@ -1538,7 +1619,305 @@ mod tests {
                 "#,
             )
             .unwrap();
-        assert_eq!(result, serde_json::json!(["P", "A"]));
+        // The rejected <section> is skipped but not pruned, so <p> still shows.
+        // The leading root is #467: an iterator yields the node it is rooted at.
+        assert_eq!(result, serde_json::json!(["DIV", "P", "A"]));
+    }
+
+    /// Issue #467: a NodeIterator starts *before* its root, so the first
+    /// nextNode() returns the root itself. Aliasing createTreeWalker silently
+    /// dropped exactly the element the iterator was rooted at.
+    #[test]
+    fn node_iterator_yields_the_root_node_first() {
+        let mut rt = setup_runtime(r#"<div id="root"><a></a></div>"#);
+        let result = rt
+            .evaluate(
+                r#"
+                const root = document.getElementById('root');
+                const it = document.createNodeIterator(root, NodeFilter.SHOW_ELEMENT);
+                const seen = [];
+                let node;
+                while ((node = it.nextNode())) seen.push(node.tagName);
+                return seen;
+                "#,
+            )
+            .unwrap();
+        assert_eq!(result, serde_json::json!(["DIV", "A"]));
+    }
+
+    /// Issue #467: the NodeIterator interface surface, and that TreeWalker-only
+    /// members are not exposed on it.
+    #[test]
+    fn node_iterator_exposes_its_own_interface() {
+        let mut rt = setup_runtime(r#"<div id="root"><a></a></div>"#);
+        let result = rt
+            .evaluate(
+                r#"
+                const root = document.getElementById('root');
+                const it = document.createNodeIterator(root, NodeFilter.SHOW_ELEMENT);
+                const before = [it.referenceNode === root, it.pointerBeforeReferenceNode];
+                it.nextNode();
+                return [
+                    before,
+                    typeof it.detach,
+                    it.detach() === undefined,
+                    typeof it.previousNode,
+                    it.root === root,
+                    it.whatToShow,
+                    // TreeWalker-only members must not leak onto a NodeIterator.
+                    typeof it.currentNode,
+                    typeof it.firstChild,
+                    typeof it.parentNode,
+                    // The pointer advanced past the root it just returned.
+                    [it.referenceNode.tagName, it.pointerBeforeReferenceNode],
+                ];
+                "#,
+            )
+            .unwrap();
+        assert_eq!(
+            result,
+            serde_json::json!([
+                [true, true],
+                "function",
+                true,
+                "function",
+                true,
+                1,
+                "undefined",
+                "undefined",
+                "undefined",
+                ["DIV", false]
+            ])
+        );
+    }
+
+    /// Issue #467: previousNode() retraces the iterator, and the root is the
+    /// last node it yields going backwards.
+    #[test]
+    fn node_iterator_previous_node_retraces_the_walk() {
+        let mut rt = setup_runtime(r#"<div id="root"><a><b></b></a><c></c></div>"#);
+        let result = rt
+            .evaluate(
+                r#"
+                const root = document.getElementById('root');
+                const it = document.createNodeIterator(root, NodeFilter.SHOW_ELEMENT);
+                const forward = [];
+                let node;
+                while ((node = it.nextNode())) forward.push(node.tagName);
+                const backward = [];
+                while ((node = it.previousNode())) backward.push(node.tagName);
+                return [forward, backward];
+                "#,
+            )
+            .unwrap();
+        // Forward ends on <c>; going back re-yields <c> (the pointer sits after
+        // it), then the rest in reverse, root included.
+        assert_eq!(
+            result,
+            serde_json::json!([
+                ["DIV", "A", "B", "C"],
+                ["C", "B", "A", "DIV"]
+            ])
+        );
+    }
+
+    /// Issue #463: `<template>` contents are parsed into the node's
+    /// `template_contents` document, but no op exposed it, so `.content` handed
+    /// back a fabricated empty fragment and the parsed markup was unreachable.
+    #[test]
+    fn template_content_exposes_parsed_markup() {
+        let mut rt = setup_runtime(
+            r#"<body><template id="t"><p class="row">a</p><p class="row">b</p></template></body>"#,
+        );
+        let result = rt
+            .evaluate(
+                r#"
+                const t = document.getElementById('t');
+                return [
+                    t.content.childNodes.length,
+                    t.content.querySelectorAll('.row').length,
+                    t.content.firstElementChild.textContent,
+                    t.innerHTML,
+                    t.content.nodeType,
+                    t.content instanceof DocumentFragment,
+                    // Identity is stable: frameworks stash `.content` and reuse it.
+                    t.content === t.content,
+                    // The children stay off the element itself, per the HTML spec.
+                    t.childNodes.length,
+                ];
+                "#,
+            )
+            .unwrap();
+        assert_eq!(
+            result,
+            serde_json::json!([
+                2,
+                2,
+                "a",
+                r#"<p class="row">a</p><p class="row">b</p>"#,
+                11,
+                true,
+                true,
+                0
+            ])
+        );
+    }
+
+    /// Issue #463: the same must hold for a template that arrives via innerHTML
+    /// rather than the initial document parse — that is how most frameworks
+    /// inject templates.
+    #[test]
+    fn template_content_works_for_templates_added_via_inner_html() {
+        let mut rt = setup_runtime(r#"<body><div id="host"></div></body>"#);
+        let result = rt
+            .evaluate(
+                r#"
+                const host = document.getElementById('host');
+                host.innerHTML = '<template id="t2"><li class="item">x</li></template>';
+                const t = document.getElementById('t2');
+                const stamped = t.content.cloneNode(true);
+                host.appendChild(stamped);
+                return [
+                    t.content.childNodes.length,
+                    t.content.querySelector('.item').textContent,
+                    host.querySelectorAll('li.item').length,
+                ];
+                "#,
+            )
+            .unwrap();
+        // cloneNode(true) of the content is the canonical stamping idiom.
+        assert_eq!(result, serde_json::json!([1, "x", 1]));
+    }
+
+    /// Issue #463: a template built with createElement has no parsed contents,
+    /// so `.content` must allocate a backing fragment on demand and round-trip
+    /// through innerHTML.
+    #[test]
+    fn template_content_round_trips_for_created_templates() {
+        let mut rt = setup_runtime(r#"<body></body>"#);
+        let result = rt
+            .evaluate(
+                r#"
+                const t = document.createElement('template');
+                t.innerHTML = '<span class="s">hi</span>';
+                return [
+                    t.content.childNodes.length,
+                    t.content.querySelector('.s').textContent,
+                    t.innerHTML,
+                    t.childNodes.length,
+                ];
+                "#,
+            )
+            .unwrap();
+        assert_eq!(
+            result,
+            serde_json::json!([1, "hi", r#"<span class="s">hi</span>"#, 0])
+        );
+    }
+
+    /// Issue #463: serializing a `<template>` must emit its contents, or the
+    /// markup silently disappears from outerHTML/innerHTML round-trips — and
+    /// `cloneNode(true)`, which round-trips through outer_html, yields an empty
+    /// template.
+    #[test]
+    fn template_contents_survive_serialization_and_clone() {
+        let mut rt = setup_runtime(
+            r#"<body><template id="t"><li class="item">x</li></template></body>"#,
+        );
+        let result = rt
+            .evaluate(
+                r#"
+                const t = document.getElementById('t');
+                const clone = t.cloneNode(true);
+                return [
+                    t.outerHTML,
+                    document.body.innerHTML,
+                    clone.content.childNodes.length,
+                    clone.content.querySelector('.item').textContent,
+                    // The clone's contents are its own, not shared with the original.
+                    (clone.content.firstElementChild === t.content.firstElementChild),
+                ];
+                "#,
+            )
+            .unwrap();
+        let expected = r#"<template id="t"><li class="item">x</li></template>"#;
+        assert_eq!(
+            result,
+            serde_json::json!([expected, expected, 1, "x", false])
+        );
+    }
+
+    /// Issue #468: window.scrollTo/scrollBy/scroll were no-op stubs, so the
+    /// dominant infinite-scroll idiom never advanced the page offset.
+    #[test]
+    fn window_scroll_methods_move_the_page_offset() {
+        let mut rt = setup_runtime(r#"<html><body><div id="d"></div></body></html>"#);
+        let result = rt
+            .evaluate(
+                r#"
+                const scrolled = window.scrollTo(0, 500);
+                const afterTo = [window.scrollX, window.scrollY];
+                window.scrollBy(0, 200);
+                const afterBy = [window.pageXOffset, window.pageYOffset];
+                window.scrollTo({ left: 10, top: 40 });
+                const afterOptions = [window.scrollX, window.scrollY];
+                window.scroll(5, 5);
+                const afterScroll = [window.scrollX, window.scrollY];
+                // Negative offsets clamp to 0, as they do for elements.
+                window.scrollTo(0, -100);
+                return [afterTo, afterBy, afterOptions, afterScroll, window.scrollY];
+                "#,
+            )
+            .unwrap();
+        assert_eq!(
+            result,
+            serde_json::json!([[0, 500], [0, 700], [10, 40], [5, 5], 0])
+        );
+    }
+
+    /// Issue #468: the page offset is one value, readable and writable through
+    /// either `window.scrollY` or `document.scrollingElement.scrollTop`.
+    #[test]
+    fn window_scroll_offset_is_shared_with_the_scrolling_element() {
+        let mut rt = setup_runtime(r#"<html><body><div id="d"></div></body></html>"#);
+        let result = rt
+            .evaluate(
+                r#"
+                const isDocEl = document.scrollingElement === document.documentElement;
+                window.scrollTo(0, 300);
+                // Written through the window, read through the element...
+                const viaElement = document.scrollingElement.scrollTop;
+                // ...and the reverse.
+                document.scrollingElement.scrollTop = 90;
+                return [isDocEl, viaElement, window.scrollY, window.pageYOffset];
+                "#,
+            )
+            .unwrap();
+        assert_eq!(result, serde_json::json!([true, 300, 90, 90]));
+    }
+
+    /// Issue #468: a scroll event must reach listeners on both the window and
+    /// the document — that is the signal lazy loaders wait for.
+    #[tokio::test(flavor = "current_thread")]
+    async fn window_scroll_fires_a_scroll_event() {
+        let mut rt = setup_runtime(r#"<html><body><div id="d"></div></body></html>"#);
+        let result = rt
+            .evaluate_for_cdp(
+                r#"
+                new Promise(resolve => {
+                    let win = 0, doc = 0;
+                    window.addEventListener('scroll', () => win++);
+                    document.addEventListener('scroll', () => doc++);
+                    window.scrollBy(0, 400);
+                    setTimeout(() => resolve([win, doc, window.scrollY]), 5);
+                })
+                "#,
+                true,
+                true,
+            )
+            .await
+            .unwrap();
+        assert_eq!(result.value.unwrap(), serde_json::json!([1, 1, 400]));
     }
 
     /// Issue #469: FILTER_SKIP leaves a skipped node's children eligible, so