2.0.0-rc.8 combined isPending reports false while both individual reads report true during an optimistic update

[GabbeV]

Describe the bug

Two async memos depend on an optimistic value. During the optimistic update, isPending(a) and isPending(b) both report true, but isPending(() => [a(), b()]) reports false.

Your Example Website or App

https://s.olid.uk/id/TRcpvUCSQ8KDmHLCmzE5JQ

Steps to Reproduce the Bug or Issue

1. Wait for initialization.
2. Click Run.
3. A and B both report pending, while Combined reports false.
4. Once A resolves, Combined becomes true while B is still loading.

Expected behavior

The combined indicator should report true while either of its reads is pending.

Screenshots or Videos

No response

Platform

• OS: Windows
• Browser: Firefox

Additional context

Replacing the optimistic value and action with an ordinary signal update makes the indicators work correctly.

Possibly related to #3379, but this uses two async siblings read by one isPending call rather than an async chain. It still reproduces after #3400.

Reproduced on next at 14ded24 and #3405 at 323f9ad, in development and production.

[ryansolid]

Confirmed and fixed in #3425 — thanks, the sibling shape was the key.

Root cause: the combined effect computes true at the write, but its run is withheld because it lands on the held value lane. Its two companion child lanes (isPending's pending signals for a and b) are siblings and merge; the parent lane's next notification then hit a shortcut in assignOrMergeLane — "subscriber's lane was merged, take the source lane" — that skipped the parent/child check and moved the effect onto the held parent, where its verdict waited on the async it reports. isPending(a) and isPending(b) each have a single child lane and never reach that path, which is why they flip immediately.

Fix: the shortcut is gone; a merged lane follows to its root and goes through the same parent/child logic. Pinned with your timeline, including the revert phase (there all three clear together when the held reload commits — same as a plain signal write).

— Claude via Cursor

[ryansolid]

Fixed in 1ce0f85 (#3425), merged to next — ships in the next rc.

The combined isPending(() => [a(), b()]) effect computed true at the write; its run was withheld because it landed on the held parent lane. In assignOrMergeLane, a's child lane was kept by the parent/child carve-out, b's child lane merged with it as a sibling, and b's parent-lane notification then hit a "subscriber's lane was merged → replace with the source lane" shortcut that skipped the parent/child check. The individual probes each have a single child lane and never reach that path, which is why they disagreed. The shortcut is gone: a merged lane is followed to its root like any other and goes through the same parent/child logic. Pinned through the revert as well.

— Claude via Cursor