mirror of https://github.com/block/buzz.git synced 2026-08-18 06:50:31 +02:00

Files

T

npub1223z34hd7vtwc6qj4s7flsxkj644nlre2nthu7lrrmkumhu3xddsrx9r6wandTaylor Ho 6f6d2320ed fix(timeline): hold first-load viewport at the bottom while the document settles

Root cause: the init pin was one-shot. On first load the virtualizer paints
with ESTIMATED row heights and the deferred snapshot streams in over several
commits, so the true bottom keeps moving AFTER the single init scrollToBottom.
Nothing re-pinned: the new-message branch only fires on a new latest-message
key (the bottom message is already present during initial fill), so the viewport
ended up anchored where the estimated bottom was and the rest filled in below as
heights resolved — exactly the 'doesn't stay attached to the bottom' tho saw.

Fix: a settle-pin layout effect re-anchors to the bottom whenever the
virtualizer's getTotalSize() changes (estimate→measured heights, streaming rows,
container resize) — but ONLY while still pinned (stickToBottomRef) and not
chasing a deep-link, so a user who scrolled up is never yanked back down. Guarded
on a real size change (lastPinnedTotalSizeRef) so the pin's own scroll-induced
re-render can't loop; scrollToBottom records the pinned size to avoid a
redundant double-fire.

Distinct from the mid-scroll prepend retention (a8b21597) and layered cleanly on
the scrollMargin work + existing sticky-bottom autoscroll — no regression to
either. Main timeline only; thread pane untouched. Two DOM tests lock in: re-pin
on size growth while pinned, and NO re-pin after the user scrolls away.

Co-authored-by: Taylor Ho <taylorkmho@gmail.com>
Signed-off-by: Taylor Ho <taylorkmho@gmail.com>

2026-06-15 21:55:48 -07:00