Merge pull request #4806 from nesquena/stage/4800-stream-perf

Release v0.51.613 — eliminate O(DOM) streaming render freeze on long answers (#4800)

Author: nesquena-hermes

CHANGELOG.md

@@ -3,6 +3,12 @@
 
 ## [Unreleased]
 
+## [v0.51.613] — 2026-06-23 — Release VT (eliminate O(DOM) streaming render freeze on long answers)
+
+### Fixed
+
+- **Long agent answers no longer progressively freeze the UI while streaming.** On answers of 30,000+ characters (5,000+ DOM nodes), every streamed token ran a full-DOM link-sanitization scan, so render cost grew with the answer size and the tab eventually froze (and could crash the renderer). Link URL-scheme validation now happens inline as each DOM node is created (via a wrapped streaming-markdown renderer), eliminating the per-token full-DOM rescan, and the streaming render now caches its parse result to skip redundant re-parsing. Dangerous URL schemes (`javascript:`, `data:`, `vbscript:`, etc.) are still rejected on links and images exactly as before, and the final settled render still runs the full sanitizer. Thanks @wlknight. (#4800)
+
 ## [v0.51.612] — 2026-06-23 — Release VS (background-tab notifications + profile default workspace on fresh sessions)
 
 ### Fixed

static/messages.js

@@ -3070,7 +3070,7 @@ function attachLiveStream(activeSid, streamId, uploaded=[], options={}){
     _smdWrittenLen=0;
     _smdWrittenText='';
     if(!window.smd){_smdParser=null;return;}
-    const baseRenderer=fade ? _streamFadeRenderer(el) : window.smd.default_renderer(el);
+    const baseRenderer=fade ? _streamFadeRenderer(el) : _safeSmdRenderer(el);
     const renderer=_smdRendererWithoutUnderscoreEmphasis(baseRenderer);
     _smdParser=window.smd.parser(renderer);
   }
@@ -3141,9 +3141,9 @@ function attachLiveStream(activeSid, streamId, uploaded=[], options={}){
     try{window.smd.parser_write(_smdParser,delta);}catch(_){}
     _smdWrittenLen=displayText.length;
     _smdWrittenText=displayText;
-    // streaming-markdown does NOT sanitize URL schemes. The default live path
-    // scans after writes; fade mode blocks unsafe href/src in its renderer.set_attr.
-    if(assistantBody&&!fade){_sanitizeSmdLinks(assistantBody);}
+    // URL scheme safety is handled by the renderer's set_attr hook
+    // (_safeSmdRenderer or _streamFadeRenderer), applied inline as smd
+    // creates each DOM node — no post-hoc full-DOM scan needed.
     _scheduleStreamingKatex();
   }
   // Allowed URL schemes for anchors and images rendered from agent-streamed markdown.
@@ -3314,6 +3314,36 @@ function attachLiveStream(activeSid, streamId, uploaded=[], options={}){
     };
     return renderer;
   }
+  // Safe renderer: wraps default_renderer with a set_attr hook that validates
+  // href/src URL schemes inline — no post-hoc DOM-wide querySelectorAll needed.
+  // Unlike _streamFadeRenderer, this does NOT wrap add_text, so smd adds new
+  // DOM nodes as plain text nodes (no animation spans). Used on the non-fade
+  // streaming path to eliminate _sanitizeSmdLinks(assistantBody) O(DOM) scans
+  // on every token event (#WebUI-perf).
+  function _safeSmdRenderer(el){
+    const renderer=window.smd.default_renderer(el);
+    const baseSetAttr=renderer.set_attr;
+    renderer.set_attr=(data,attr,value)=>{
+      const isHref=window.smd&&attr===window.smd.HREF;
+      const isSrc=window.smd&&attr===window.smd.SRC;
+      const safeUrl=isSrc?_SMD_SAFE_IMG_URL_RE:_SMD_SAFE_URL_RE;
+      if(isHref&&/^(file|workspace|session):\/\//i.test(String(value||''))){
+        baseSetAttr(data,attr,_smdLinkHref(value));
+        if(/^session:\/\//i.test(String(value||''))){
+          const node=data&&data.nodes&&data.nodes[data.index];
+          if(node&&node.classList) node.classList.add('session-link');
+        }
+        return;
+      }
+      if((isHref||isSrc)&&!safeUrl.test(String(value||''))){
+        const node=data&&data.nodes&&data.nodes[data.index];
+        if(node&&node.setAttribute) node.setAttribute('data-blocked-scheme','1');
+        return;
+      }
+      baseSetAttr(data,attr,value);
+    };
+    return renderer;
+  }
   function _streamFadeWordCountOf(text){
     const m=String(text||'').match(/\S+/g);
     return m?m.length:0;
@@ -3784,7 +3814,19 @@ function attachLiveStream(activeSid, streamId, uploaded=[], options={}){
   }
 
   let _lastRenderMs=0;
-  function _scheduleRender(){
+  // Parse-result cache: _scheduleRender can accept a pre-computed _parseStreamState()
+  // from the token event handler, avoiding a duplicate O(n) scan inside _doRender
+  // when the rAF fires before the next token arrives.
+  let _cachedParsed=null;
+  let _cachedParsedText='';
+  let _cachedParsedReasoning='';
+  function _scheduleRender(parsed){
+    // If caller provides a pre-computed parse result, cache it for _doRender.
+    if(parsed){
+      _cachedParsed=parsed;
+      _cachedParsedText=assistantText;
+      _cachedParsedReasoning=liveReasoningText;
+    }
     if(_renderPending) return;
     if(_streamFinalized) return; // Bug A: don't schedule new rAF after stream finalized
     _renderPending=true;
@@ -3803,7 +3845,8 @@ function attachLiveStream(activeSid, streamId, uploaded=[], options={}){
       // Guard: a pending setTimeout+rAF can outlive stream finalization.
       if(_streamFinalized) return;
       _lastRenderMs=performance.now();
-      const parsed=_parseStreamState();
+      const parsed=_cachedParsed&&_cachedParsedText===assistantText&&_cachedParsedReasoning===liveReasoningText ? _cachedParsed : _parseStreamState();
+      _cachedParsed=null;
       _renderLiveThinking(parsed);
       if(assistantBody){
         const displayText = segmentStart===0
@@ -3898,7 +3941,7 @@ function attachLiveStream(activeSid, streamId, uploaded=[], options={}){
       const parsed=_parseStreamState();
       if(_freshSegment) appendThinking('', _liveThinkingPlacement());
       if(String((parsed&&parsed.displayText)||'').trim()||assistantRow) ensureAssistantRow();
-      _scheduleRender();
+      _scheduleRender(parsed);
     });
 
     source.addEventListener('interim_assistant',e=>{

tests/test_stable_assistant_turn_anchor_registry.py

@@ -1301,7 +1301,7 @@ def test_slice6_live_shadow_feed_wires_anchor_scene_for_visible_order_handoff():
         assert f"_applyToAnchor('{event_name}'" in _event_listener_body(src, event_name)
 
     token_body = _event_listener_body(src, "token")
-    assert "_scheduleRender();" in token_body
+    assert "_scheduleRender(" in token_body
     assert "function _upsertAnchorProcessProse" in src
     assert "_upsertAnchorProcessProse(displayText" in src
     reasoning_body = _event_listener_body(src, "reasoning")

tests/test_streaming_markdown.py

@@ -197,10 +197,18 @@ def test_smd_new_parser_resets_written_len(self):
             "_smdWrittenLen=0" in fn or "_smdWrittenLen = 0" in fn
         ), "_smdNewParser must reset _smdWrittenLen to 0"
 
-    def test_smd_new_parser_calls_default_renderer(self):
+    def test_smd_new_parser_calls_safe_renderer(self):
         fn = extract_fn(MESSAGES_JS, "_smdNewParser")
-        assert fn and "default_renderer" in fn, (
-            "_smdNewParser must call smd.default_renderer() to create a renderer"
+        assert fn and (
+            "_safeSmdRenderer(" in fn or "_streamFadeRenderer(" in fn
+        ), (
+            "_smdNewParser must use _safeSmdRenderer or _streamFadeRenderer "
+            "so URL scheme safety is applied inline via set_attr hook"
+        )
+        # Verify _safeSmdRenderer itself uses smd.default_renderer internally
+        safefn = extract_fn(MESSAGES_JS, "_safeSmdRenderer")
+        assert safefn and "default_renderer" in safefn, (
+            "_safeSmdRenderer must call smd.default_renderer() to create the base renderer"
         )
 
     def test_smd_new_parser_calls_parser(self):
@@ -706,16 +714,41 @@ def test_file_anchor_rewrite_helper_exists(self):
         assert "_smdFileHref" in MESSAGES_JS
         assert "api/media?path=" in MESSAGES_JS
 
-    def test_sanitize_called_after_smd_write(self):
-        # _smdWrite must invoke _sanitizeSmdLinks on assistantBody after feeding the parser,
-        # so anchors/images created mid-stream get their javascript:/data:/vbscript:
-        # hrefs/srcs stripped before the user can click them.
-        fn = extract_fn(MESSAGES_JS, "_smdWrite")
-        assert fn, "_smdWrite function not found"
-        assert "_sanitizeSmdLinks" in fn, (
-            "_smdWrite must call _sanitizeSmdLinks(assistantBody) after parser_write "
-            "so unsafe URL schemes are stripped from newly-added anchors/images "
-            "before the user can click them"
+    def test_url_safety_via_renderer_set_attr(self):
+        # URL scheme safety is now enforced inline by the renderer's set_attr
+        # hook (_safeSmdRenderer or _streamFadeRenderer) as smd creates each
+        # DOM node, eliminating the post-hoc _sanitizeSmdLinks O(DOM) scan
+        # per token that caused progressive streaming freeze on long answers.
+        safefn = extract_fn(MESSAGES_JS, "_safeSmdRenderer")
+        assert safefn, "_safeSmdRenderer must exist for URL safety on the non-fade path"
+        assert "set_attr" in safefn, (
+            "_safeSmdRenderer must override set_attr to validate href/src inline"
+        )
+        assert "_SMD_SAFE_URL_RE" in safefn, (
+            "_safeSmdRenderer set_attr must use _SMD_SAFE_URL_RE for href safety"
+        )
+        assert "_SMD_SAFE_IMG_URL_RE" in safefn, (
+            "_safeSmdRenderer set_attr must use _SMD_SAFE_IMG_URL_RE for src safety"
+        )
+        assert "data-blocked-scheme" in safefn, (
+            "_safeSmdRenderer set_attr must set data-blocked-scheme on unsafe URLs"
+        )
+        # _safeSmdRenderer algorithm must be the same as the proven
+        # _streamFadeRenderer set_attr (fade path has been in production).
+        fadefn = extract_fn(MESSAGES_JS, "_streamFadeRenderer")
+        assert fadefn and "set_attr" in fadefn, "_streamFadeRenderer must also have set_attr"
+        # Both renderers go through _smdRendererWithoutUnderscoreEmphasis so
+        # the set_attr hook is part of the final parser — verify the call chain.
+        newparser = extract_fn(MESSAGES_JS, "_smdNewParser")
+        assert newparser and "fade ? _streamFadeRenderer(el) : _safeSmdRenderer(el)" in newparser, (
+            "_smdNewParser must route non-fade to _safeSmdRenderer and fade to _streamFadeRenderer"
+        )
+        # _sanitizeSmdLinks is still defined and used at parser_end as a final
+        # safety net — not removed, just no longer called on every token.
+        assert "_sanitizeSmdLinks" in MESSAGES_JS, "_sanitizeSmdLinks must still exist"
+        endfn = extract_fn(MESSAGES_JS, "_smdEndParser")
+        assert endfn and "_sanitizeSmdLinks" in endfn, (
+            "_smdEndParser must still call _sanitizeSmdLinks as a final safety net"
         )
 
     def test_sanitize_called_at_parser_end(self):

tests/test_streaming_race_fix.py

@@ -42,7 +42,7 @@ def test_pending_raf_handle_declared(self):
 
     def test_schedule_render_guards_on_stream_finalized(self):
         src = read('static/messages.js')
-        m = re.search(r'function _scheduleRender\(\)\{.*?\n  \}', src, re.DOTALL)
+        m = re.search(r'function _scheduleRender\([^)]*\)\{.*?\n  \}', src, re.DOTALL)
         assert m, "_scheduleRender not found"
         fn = m.group(0)
         assert '_streamFinalized' in fn, (