Release XR (v0.51.662): ignore DOM-wipe scroll clamp during render (#4934)

Release XR (v0.51.662): ignore DOM-wipe scroll clamp during render (#4934)

CHANGELOG.md

@@ -3,6 +3,12 @@
 
 ## [Unreleased]
 
+## [v0.51.662] — 2026-06-25 — Release XR (live replies keep following during transcript rebuilds)
+
+### Fixed
+
+- **A streaming reply no longer stops following (or appears to jump backward) when the transcript rebuilds mid-stream.** `renderMessages` rebuilds the message pane by clearing it (`innerHTML=''`), and during that brief gap the scroll container can collapse and the browser clamps `scrollTop` to 0, emitting a scroll event. The scroll listener was interpreting that render artifact as a deliberate user scroll-up and unpinning the reader, so the live reply stopped auto-following. The DOM-wipe scroll is now marked programmatic and ignored; genuine user scroll-up (wheel/touch) during a render is still detected and still unpins. Thanks @allenliang2022. (#4934)
+
 ## [v0.51.661] — 2026-06-25 — Release XQ (no more fabricated multi-hour processing time)
 
 ### Fixed

static/ui.js

@@ -11316,6 +11316,13 @@ function renderMessages(options){
   // Mobile scroll-jank fix: temporarily disable overflow-anchor so Chromium
   // cannot re-anchor to the topmost row during the DOM wipe-and-rebuild gap.
   if(window._fixMobileScrollJank) window._fixMobileScrollJank();
+  // The DOM wipe can briefly collapse #msgInner to zero height, causing the
+  // browser to clamp #messages.scrollTop to 0 and emit a scroll event.  That
+  // event is a render artifact, not user intent; if the scroll listener sees it
+  // with _programmaticScroll=false, it marks the reader manually unpinned and
+  // the live reply stops following / appears to jump backward.
+  _programmaticScroll=true;
+  _programmaticScrollSetAt=performance.now();
   inner.innerHTML='';
   const compressionNode=compressionState?_compressionCardsNode(compressionState):null;
   const {message:referenceMessage, rawIdx:referenceMessageRawIdx}=_latestCompressionReferenceMessage(
@@ -12524,6 +12531,7 @@ function renderMessages(options){
   }
   _updateMessageVirtualMeasurements(renderVisWithIdx, renderVisibleIdxs, virtualWindow);
   _recycleStash.clear();
+  if(typeof _deferClearProgrammaticScroll==='function') _deferClearProgrammaticScroll(160);
 }
 
 function _toolDisplayName(tc){

tests/test_issue4856_android_scroll_regression.py

@@ -83,6 +83,27 @@ def test_rebuild_path_calls_fix_before_wipe():
     )
 
 
+def test_rebuild_path_marks_dom_wipe_scroll_as_programmatic():
+    # During innerHTML='' the scroller can transiently collapse to clientHeight
+    # and clamp scrollTop to 0. That browser event must be suppressed as
+    # programmatic; otherwise the scroll listener treats it as user upward
+    # intent and disables live auto-follow.
+    fix_idx = UI_JS.find("window._fixMobileScrollJank()")
+    assert fix_idx != -1, "renderMessages() guard call not found"
+    wipe_idx = UI_JS.find("innerHTML=''", fix_idx)
+    assert wipe_idx != -1, "innerHTML='' not found after _fixMobileScrollJank()"
+    window = UI_JS[fix_idx:wipe_idx]
+    assert "_programmaticScroll=true" in window, (
+        "renderMessages() must mark the DOM wipe/rebuild scroll event as "
+        "programmatic before innerHTML='' can clamp scrollTop."
+    )
+    assert "_programmaticScrollSetAt=performance.now()" in window
+    assert UI_JS.find("_deferClearProgrammaticScroll(160)", wipe_idx) != -1, (
+        "renderMessages() must clear the programmatic-scroll suppression after "
+        "the rebuild/post-render paint window."
+    )
+
+
 def test_streaming_tick_calls_fix_before_dom_writes():
     # The streaming render tick in messages.js must call _fixMobileScrollJank()
     # before _lastRenderMs=performance.now() so anchor suppression covers every