The base rule declared --scroll-shadow-size as var(--scroll-shadow-size,
48px) — a self-referential cycle, which computes to invalid and silently
resolves every mask below to none. The old ScrollShadow component masked
the bug by setting the variable inline; the hook-based path (the chat
scroller) sets only the data attributes, so the chat lost its top and
bottom scroll fades. A concrete 48px default restores them; inline
overrides from the component still win.
The animated maintainScrollAtEnd also ran outside streaming, so opening
a historical session glided visibly through the whole conversation as
late row measurements corrected the end position, and an in-flight glide
could supersede explicit navigation. Corrections are instant unless the
session is actively working.
Returning to the bottom during a stream also lands on the end as of that
moment, and the list's own follow may not have re-armed after the user's
earlier gestures — the queued-send jump then fell behind the growing
reply. goToBottom now re-asserts the edge a few times (150/400/800ms)
until it holds; a new user gesture cancels the window.
During a long stream the follow scroll masks the entrance entirely, so
the blur was invisible where it streamed and only a cost risk on weak
GPUs. Keep the compositor-only fade and rise (and the same-tick cascade)
— those are effectively free and still show on short replies where no
scroll motion is involved.
The 140ms fade read as a pop next to the scroll glide. The reveal now
runs 320ms on an ease-out-quint curve with a 10px rise and a touch of
blur resolving — paint-only properties throughout, so row measurement is
untouched — and blocks committed in the same tick cascade with a 55ms
stagger instead of appearing together. Reduced motion still disables the
whole thing.
Retain detached Markdown DOM for eight recent sessions so medium-frequency session switching can restore more first-screen content without reparsing it.
Limit each session to four cached parts, reducing the worst-case retained fragment count from 36 to 32 while preventing long sessions from dominating the cache.
Keep sidebar rows independent from the active sync directory by moving export history loading behind an explicit-directory command.
Make active-project selection preserve project topology and remove activeProjectId from mounted group render props. Reuse a per-session-array ID index so subscription snapshots do not repeatedly scan session lists.
Batch global session, status, ordering, and activity timing mutations at the existing directory event boundary so large subagent bursts publish each owner once.
Maintain active session roots, children, and directory buckets in the global store, reuse them in Sidebar projections, and avoid rebuilding live aggregates and structural data for unrelated renders while preserving authoritative ordering reconciliation.
Skip mutation history when no settings operation is in flight and merge updates that share the same operation visibility window. This keeps stale-response reconciliation unchanged while preventing history from growing with repeated debounced updates.
Add a repeated-update regression covering an older pending load and latest-value reconciliation.
Tests: bun test src/lib/persistence.test.ts
Keep settings mutation revisions and in-flight operations behind a private tracker so stale GET and PUT responses cannot overwrite newer local settings.
Remove redundant runtime identity and numeric operation IDs. Use operation object identity while preserving runtime reset and mutation retention behavior.
Tests: bun test src/lib/persistence.test.ts
With block-level commit the open code fence grows by whole lines at the
throttle cadence, so the old reason to skip highlighting it (re-tokenizing
a growing block ~40x/sec) no longer applies: a partial fence now
highlights too, and streamed code arrives colored instead of colorizing
only when the fence closes. Fences beyond 300 lines fall back to plain
text until closed, keeping the repeated worker re-tokenization bounded.
A freshly committed block also enters with a short fade-and-rise. The
class goes on the block's children (the wrapper is display:contents and
cannot animate) and the transform is paint-only, so virtualized row
measurement is unaffected; reduced motion disables it.
Verified over CDP on a production build: streamed code inside a still-
open fence renders highlighted, and the code text's x-position is
identical during and after the stream.
Token-by-token streaming mutates the trailing paragraph in place on
every tick: words rewrap, the last line jitters, and the whole reply
reads as flicker. Streamed text now commits only up to the last complete
line — prose arrives a paragraph at a time (a markdown paragraph is one
logical line), code fences reveal line by line, tables row by row — and
a shown block never changes again. A paragraph that runs long without a
newline releases at the last sentence (then word) boundary so the stream
never stalls. Applies to assistant text and reasoning; tool output keeps
its raw tail.
With growth arriving in block steps, the end follow switches to the
list's animated mode so each step is a glide — reveal and scroll read as
one continuous motion. The gesture opt-out now measures at-end from the
live list state instead of the cached flag, which the animated glide
deliberately leaves stale while trailing the edge; without that, a drag
during a glide could leave the scroll-to-bottom pill unshown.
Measured: end-following holds at distance 0 for the whole stream, and
the mobile drag opt-out shows the pill in three of three runs. The
continuous glide costs ~15% more main-thread time per streamed character
than the instant follow — the price of the motion.
Three first-render polish issues:
Code blocks jumped at end-of-stream: streaming defers the per-line
line-number markup, so the finished decorate pass inserted the gutter
column and shifted every code line right. The gutter's horizontal
footprint is now reserved with CSS while the markup is deferred, so the
final pass only fills in numbers and colors.
File references were verified twice per render and re-verified with the
wrong directory: the annotation pass's own DOM writes re-triggered its
MutationObserver, and the pass also ran before the effective directory
resolved — issuing stat probes under an empty directory and a second
time under the real one. The observer now ignores the pass's own
mutations and annotation waits for a resolved directory.
Content replacing the hydration skeleton popped in: it now plays a
one-shot 180ms fade (reduced-motion aware); cached session switches
never carry the class and stay instant. Also removed the dead
disableStaging prop and its unused pendingRevealWork threading.
Verified on a production build over CDP: a streamed code block's text
keeps its exact x-position across end-of-stream, and stat probes for a
message with file mentions are unique per path with the directory header
always present. The hydration fade path could not be exercised in the
harness (the live event stream pre-populates parts in a single-project
environment) and needs an eyeball check on a cold multi-project open.
Sending from mid-history positioned the new turn with an animated
scrollToIndex; over a long distance the smooth scroll gets cancelled by
rows mounting and measuring along the way — a weak scroll attempt that
dies partway, leaving the reader where they were. Park instantly when
the viewport was not at the end at send time; the short smooth park
remains for sends from the live edge.
Sending while the agent is working queues the message without creating a
user row, so the anchor-arming send path had nothing to claim and the
viewport stayed parked mid-history. Queued sends now jump straight to
the live edge (instant, so end following re-latches immediately).
Rail selection also switches from smooth to instant scrolling: a long
smooth scroll through the virtualized timeline gets cancelled by row
remounts and lands mid-way or on the wrong message, while a teleport
always arrives.
3.3.x makes maintainScrollAtEnd follow content growth on its own — a
tail row growing in place included — which is exactly what the manual
totalSize correction existed for. Delete that correction (the totalSize
listener now only drives the anchored-turn glide) and pick up 3.3.x's
measurement batching, prepend-flash fixes, and web programmatic-scroll
fixes. Opt the explicit maintainScrollAtEnd config into footerLayout per
the 3.1.1 guidance.
The library's own released-on-user-scroll heuristic proved unreliable
one run in three against synthetic touch, so the gesture state machine
stays authoritative: while a real gesture owns the scroll, the list's
end pinning is switched off through a threaded endPinningReleased prop
and re-engages when the user returns to the end.
Validated with the CDP battery on a production build: stream follow
stays at distance 0, mobile drag releases with the pill shown in three
of three runs, resize oscillation stays at the reduced level, the rail
reaches the last turn, and profiled streaming cost per rendered
character matches the tuned 3.2.0 numbers.
A slow edge drag settles the resize detector repeatedly, and every settle
replayed the instant return to the live edge — reading as exactly the
jumping the resize suspension removes. Stay where the reader is; geometry
changed, and the next explicit action or stream correction re-engages the
edge as usual.
A pinned viewport shook during window resizes while a scrolled-up one
stayed calm: both pinning mechanisms — the growth corrections and the
list's maintain-scroll-at-end — re-assert the end against rows that are
still re-measuring, fighting the size compensation that keeps the free
case stable. While the width is actively changing, stand both down and
hold the reading position the same way the free path does, then re-assert
the live edge once with a single instant write after the resize settles
(only when the viewport was still following).
Both scrollToTurnId and scrollToMessageId hard-returned false when the
target was the trailing (last) turn and its element was not mounted, so
clicking the rail's last item in a long scrolled-up session did nothing.
The trailing entry is a regular list row at the end of the data — scroll
to its index the same way history targets are reached.
A width change re-wraps every row at once, and with size restoration off
the accumulated height delta above the viewport threw the visible content
hundreds of pixels up and down while dragging the window edge. Size
restoration must stay off in steady state — a tool result expanding in
place has to grow downward — so a ResizeObserver on the scroll node
enables maintainVisibleContentPosition size compensation only while the
width is actively changing and releases it 300ms after it settles.
Measured on a continuous 1400-800-1400 drag over a long session pinned to
the live edge: the largest per-step displacement of a visible paragraph
drops from ~680px (median of three runs) to ~200px, and the viewport
never detaches from the end.
On touch surfaces the gesture opt-out waited for the viewport to have
already left the end, but streaming corrections re-pin the viewport every
chunk, so the drag never registered: the user could not scroll away
mid-stream, the scroll-to-bottom pill (which waits for a gesture) never
appeared, and live-follow stayed armed — which also made the mobile
load-older button throw the viewport to the bottom, since the prepend's
content growth ran the end correction.
Track the finger's Y between touch events and treat a downward drag (the
touch mirror of wheel-up) as an immediate opt-out. Loading older history
also releases live follow explicitly before the prepend, covering the
case where the button is reachable without any prior scroll.
The old scroll engine needed message parts to report content growth
(onContentChange) and per-message animation lifecycle callbacks
(AnimationHandlers) so it could re-pin the viewport. The timeline list
measures growth itself now, and the replacement hook had already stubbed
the whole contract with no-ops kept only for source compatibility.
Remove it end to end: the hook exports, the container and list threading,
the ChatMessage/MessageBody signal-only effects, and every part-level
prop and call site. Expand/collapse behavior and reveal animations are
untouched — only the reporting channel goes.
New Streaming section on the Chat settings page with a checkbox that
controls whether the viewport follows new content while a response
streams. Default stays on. With it off, the anchored user message still
parks at the top on send, but no glide or end-follow correction runs and
the list's maintain-scroll-at-end stays disabled; the scroll-to-bottom
pill and session open keep scrolling explicitly.
Persisted through desktop settings like the other chat toggles (auto-save
diff, authoritative apply, sanitize), registered in settings search, and
localized in every locale.
Streaming follow scheduled a two-frame animation-frame chain per content
growth and then went through the list's programmatic scrollToEnd, whose
scroll bookkeeping schedules further frames. Together they roughly
doubled frame production for the whole stream (16 -> 35 frames/s in a
profiled streaming session), with the matching style recalc and layout
cost per frame.
In following-end mode, correct synchronously in the totalSize listener by
writing scrollTop directly — the same write path a user gesture takes —
and keep the animation-frame chain only for the anchored-turn glide.
Profiled streaming frame production returns to the previous engine's
range and peak heap drops from 164 MB to 90 MB.
The streaming tail only overlaid the actively streaming message with live
parts from the sync store. When a turn moved to its next step message, the
previous message fell back to its lagging base record, briefly dropping its
completed tool parts — remounting them and replaying their reveal
animation once the record caught up.
Overlay live parts for every message of the streaming tail via a new
useSessionPartsForMessages hook, guarded so an empty live array never
erases parts the record does have. Also key generate-effect glyphs by
index so appended text does not replay earlier characters, and keep tool
row reveal wrappers mounted unconditionally.
The streaming status and the scroll-to-bottom pill now hand off to each
other in one place above the composer: same anchor, same input-aligned
column, both in the popover glass (ring and soft shadow stay pill-only
as the interactivity cue, with the shadow on a wrapper so the glass
backdrop-filter cannot drop it after hide/show cycles). The pill is a
single full-surface button, moves to the left edge, and while the
session streams it carries a compact one-line status label instead of
the status row's animation machinery, which did not survive a 32px
chip.
Behavioral fixes along the way: a gesture shows the pill immediately
(no debounce) and hides the status in the same frame; gestures register
whenever the viewport can actually move up — an anchored short turn
kept live-follow armed and suppressed the pill entirely; the status
row sheds its in-list morph offsets (mb-6, message column) and its
inline-size container gets the glass on an inner row, since a query
container cannot shrink-wrap; its working text uses full
muted-foreground to match the pill.
The animated glide trails its target between corrections, so the
distance-based at-end signal flapped during streaming and the pill and
the status row crossfaded into overlapping half-transparent mush. The
pill now shows only in free-scrolling (a real gesture), which also
keeps the status row's visibility stable; a gesture taken after our own
movement already left the end offers the pill immediately.
Away from the live edge the floating status row fades out (it was
transparent text over scrolled history) and the scroll-to-bottom pill
takes over the activity signal with a pulsing dot. The pill moves to
the left edge — matching the status row it stands in for — and gets the
same glass treatment as the dropdowns.
The rest gap after completion doubled up: the status overlay reserved
its own end inset on top of the footer's 10vh spacer. The overlay now
floats within the spacer zone — its height feeds only the glide target
(the live line still rests above it), the list gets no extra inset, and
completion cannot shift layout because there is nothing to collapse.
The finished message's metadata lands in the slack the glide left above
the status row, restoring the old morph-in-place feel. The frozen
height hack and the opaque background are gone — the overlay only ever
covers blank spacer, not text.
The barely-visible shift at end-of-stream was the reserved status-row
inset being released when the row unmounted. The measured overlay
height now only grows during a session and resets on session switch, so
the finished message's metadata footer appears exactly where the status
row was, with zero layout movement.
The floating status/working row let streaming text show through it and,
because the scroll math did not know the overlay existed, the reveal
correction targeted the very bottom edge — the live line ended up
hidden beneath the status row ('catches up to almost-the-end'). The
overlay now has a solid background, its height is measured with a
ResizeObserver and fed to the timeline as the occluding overlay height,
so the live streaming line glides to rest just above the status row and
the list reserves matching end inset at rest.
The phase-2 reveal correction now scrolls animated; successive
corrections restart the smooth scroll from the current position, so
streaming follows as one continuous motion. User gestures interrupt the
native smooth scroll, so free-scrolling is unaffected.
Phase 2 of the anchored turn (following the live edge once the reply
outgrows the viewport) never ran: the reveal correction was triggered
only by entries-array changes, and the streaming tail grows inside a
single row without touching the array. The hook now also subscribes to
LegendList's totalSize and re-runs the same guarded correction on every
content growth.
Six reported defects, one root theme — our port diverged from the
reference semantics:
- a user gesture no longer collapses the reserved anchored end space
(that snap-to-bottom was the 'anchoring vanishes on a micro-scroll');
it only disarms the follow/anchor machinery, and anchor remeasures
after the user takes over can no longer re-position the viewport
- the anchor arms synchronously BEFORE the send reaches the store, and
the claim compares against the arm-time baseline — arming a frame
after the optimistic row committed is why anchoring 'rarely worked'
and sending from mid-history did nothing
- gestures are selective (wheel up, touch/pointer away from the end,
PageUp/Home/ArrowUp), so scrolling toward the end or clicking a row
at the live edge no longer kills follow
- 'at end' is a tight 40px band against the full content length instead
of the list's half-viewport isNearEnd, so the scroll-to-bottom pill
appears when the viewport actually leaves the end
- reaching the end re-arms follow without knocking an anchored turn out
of its mode, and the overlay scrollbar suppression follows scroll
ownership rather than the anchor's mere existence
- the status/working row moved out of the list footer to a fixed spot
above the composer: inside the list it walked down with every
streamed line, and its unmount was part of the end-of-stream jerk
Sending a message now parks that message near the top of the viewport
and streams the reply into reserved end space below it, instead of
jumping to the bottom and chasing it.
- swap @tanstack/react-virtual for @legendapp/list in the chat timeline; the
streaming tail becomes a normal list row rather than a separately rendered
block, so one component owns the scroll position
- add timelineScrollAnchoring: pure anchored-turn geometry plus the three
scroll modes (following-end / anchoring-new-turn / free-scrolling)
- replace useChatAutoFollow with useChatTimelineScroll, which opts out of
automatic movement on real gestures via a generation counter instead of the
timer windows the old implementation needed to recognise its own writes
- move the load-older button, question/permission cards, recap, status row and
bottom spacer into the list header/footer, since the list owns its container
- extract useScrollShadow so the shadows can attach to that container
maintainScrollAtEnd and maintainVisibleContentPosition replace the manual
prepend anchor-hold and the mobile quiet-window prepend deferral.
Validated: workspace type-check, lint, web build, ui tests per file.
Scroll behaviour itself is unverified and needs manual testing on web, desktop
and iOS.
The hover info card (project/branch/PR/time) added noise without pull —
the tab title plus the sidebar already cover it. The tooltip body
component, its per-hover subscriptions, and the native title attribute
are gone; right-click/menu and the status dot are untouched.
Bohdan Triapitsyn