solidjs / solid / 41254025070

78%

main: 89%

push

github

fix(signals): a guess over a question in flight keeps the frame whole, shared or split bindings (#4536) (#4558)

* fix(signals): retain query observation during optimistic rerenders

Keep an existing reader linked to a pending store-family derive when an optimistic guess reruns the reader. Add shared-binding signals and web regressions and a patch changeset.

Addresses solidjs/solid#4536.

* fix(signals): keep a guess over a question in flight off the screen until its answer, in one frame (#4536)

Builds on the contributor's guard (a reader already linked to the
derive's flight keeps observing it when a guess re-runs it):

- The guard's link check walks the derive's subscribers instead of the
  reader's dependencies. A memo reading N rows asked it per row and
  walked all N deps each time while a guess stood over a pending derive
  (quadratic: 7.9 s at 4k rows, back to next's speed now).
- A reader pending on the flight without a link (a binding that reads
  only the list, beside one that reads the filter) keeps observing it
  too, when the guess was written while the derive asked that question
  (the #4525 mark, stamped on the family with the flight's `_time`).
  Otherwise its lane pass was served the guess, left the hold, and the
  frame committed `closed` beside the open answer's rows (F-1, D-34).
- When that question's answer re-bases the guess (#4525), the lane is
  marked with the question's transaction; in that transaction's flush
  the lane's work reads its staging like the frame's own passes. A memo
  that read the re-based rows before the held filter otherwise published
  `open: [#2]` for one frame before the commit's `closed: [#2]`.

The shared-effect test now also rejects the old question beside the new
answer's rows; a split-bindings test covers both `on` postures.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015hPxMkC3jg3Uivh65gvaf2

* fix(signals): only a reader in the frame's h... (continued)

797 of 1098 branches covered (72.59%)

Branch coverage included in aggregate %.

1129 of 1367 relevant lines covered (82.59%)

29.92 hits per line