2.0.0-rc.8 latest stops updating after intermediate write restores the initial value

[GabbeV]

Describe the bug

An action sets a signal to 2, back to 0, then to 2 again. A combined expression reading latest(count) and an async memo derived from it stays at 0 / 0, even after the action and all async work finish.

Your Example Website or App

https://s.olid.uk/id/uyKDwlKMSai1F0-BofC7DA

Steps to Reproduce the Bug or Issue

1. Wait for 0 / 0.
2. Click the button.
3. Wait a few seconds. It remains 0 / 0.

Expected behavior

The expression eventually shows 2 / 2.

Screenshots or Videos

No response

Platform

• OS: Windows
• Browser: Firefox

Additional context

Removing the direct latest(count) read from the expression lets the async memo display 2 correctly. Replacing both uses of latest(count) with count() also works.

Reproduced against next at 344ed05 and #3337 at 20f3143.

[ryansolid]

Confirmed and fixed in #3392 (stacked on #3381) — thanks for the reproduction.

Root cause: Pair reads latest(count), which makes it follow the optimistic lane whenever the staged count differs from the committed one. On the pass after setCount(0) (equal to committed, so no lane) it staged Pair: 0 / 0 as a transaction-held value; on the final pass it followed the lane and committed Pair: 2 / 2 directly — but the stale held 0 / 0 was left in the slot, and the transaction's commit published it over the fresh value. A lane commit now retires the hold it supersedes. Removing latest(count) from Pair "worked" because without it Pair never followed the lane.

— Claude via Cursor

[ryansolid]

Fixed on next in 6095955 (#3392). pair had staged Pair: 0 / 0 as a transaction hold on the pass where setCount(0) equality-cancelled the latest() companion's write (no lane); on the later lane pass it direct-committed 2 / 2 but left the older hold in the slot, and the transaction's commit published the stale frame over it. A lane recompute now drops a superseded hold whether or not the node carries an override.

— Claude via Cursor