NEC-296: A word deleted from a note came back in the note property test on CI (seed 82, timing-dependent) #302

Merged

nectenda-agent merged 6 commits from worktree-nec-296-a-word-deleted-from-a-note-came-back-in into main 2026-10-17 10:40:42 +01:00

Conversation 0 Commits 6 Files changed 10 +652 -71

nectenda-agent commented 2026-10-17 02:01:21 +01:00

Collaborator

Copy link

Task: NEC-296 — https://projectron.nerchure.com/tasks/309

A note's document stopped while its seed was being hashed carried on when the hash finished: it stored a stale file record over the next session's, read the file in and reported text as adopted for a session that was gone. The next session merged against the stale base, so lines deleted since came back — the note property test's rare CI failure (seed 82, then caught with evidence at seed 71, 1 in 900 loaded soak iterations). A stopped document now drops late work: no record (persistRecord, beginWrite), no read-in, no adopted text, no scheduled write, and a write re-checks liveness after its last read (SAFE-A44). The IndexedDB test stand-in now refuses reads and writes after destroy, as the real provider does. Also: the property test prints per-step evidence on failure, and unit-soak's failure excerpt starts at the assertion (ride-along NEC-299). Soak after the fix: 900 idle + 900 loaded, 0 failures.

Spec: docs/changes/NEC-296-deleted-word-came-back-property-test/spec.md

Also fixes NEC-299: unit-soak's failure excerpt is filled by stderr and never reaches the assertion

Changelog

Fixed a rare case where a note reconnected while its first sync was being prepared could act on stale text, which could have undone recent typing awaiting replay in an open editor.

Task: NEC-296 — https://projectron.nerchure.com/tasks/309 A note's document stopped while its seed was being hashed carried on when the hash finished: it stored a stale file record over the next session's, read the file in and reported text as adopted for a session that was gone. The next session merged against the stale base, so lines deleted since came back — the note property test's rare CI failure (seed 82, then caught with evidence at seed 71, 1 in 900 loaded soak iterations). A stopped document now drops late work: no record (persistRecord, beginWrite), no read-in, no adopted text, no scheduled write, and a write re-checks liveness after its last read (SAFE-A44). The IndexedDB test stand-in now refuses reads and writes after destroy, as the real provider does. Also: the property test prints per-step evidence on failure, and unit-soak's failure excerpt starts at the assertion (ride-along NEC-299). Soak after the fix: 900 idle + 900 loaded, 0 failures. Spec: `docs/changes/NEC-296-deleted-word-came-back-property-test/spec.md` Also fixes NEC-299: unit-soak's failure excerpt is filled by stderr and never reaches the assertion ## Changelog Fixed a rare case where a note reconnected while its first sync was being prepared could act on stale text, which could have undone recent typing awaiting replay in an open editor.

nectenda-agent added the

area:plugin

label 2026-10-17 02:01:21 +01:00

nectenda-agent added 6 commits 2026-10-17 02:01:21 +01:00

Make a failed note property sequence carry its own evidence ... 606295e9ff

Seed 82 failed once on CI with a deleted word back, and never again in
several thousand runs locally or 80 on the runner. A failing sequence now
prints every step's document, file, base, lastSeen, client ids and seed
digests in flight, and the fake store's event log, so the next one can be
read rather than guessed at. Recording takes no turn: all 150 seeds settle
identically with it on and off (client ids pinned).

Task: NEC-296

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YHSWLicoRvWzsYjkfdNKYz

Name each restart's client in the trace, and claim only what was shown ... 746c611a73

Review: the plan asked for the client id at every restart in the trace;
"has not happened again" is about CI history nobody checked, so it reads
"not reproduced"; the two non-deterministic sources are from reading the
code and say so; and the on/off check was against the file before the
change, which the spec now records. Re-checked: 150 seeds identical.

Task: NEC-296

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YHSWLicoRvWzsYjkfdNKYz

Print a failed soak iteration from its assertion, not its first error ... 2f9eabe4f4

The excerpt took the first three Error: lines and capped at 120. A test
that logs errors to stderr filled it before the assertion, and the
iteration's own log is lost with the runner: soak run 1483's one failure
in 600 left nothing to read. Now from the first AssertionError, 600
lines; the old grep when there is none.

Task: NEC-296
Also-fixes: NEC-299

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YHSWLicoRvWzsYjkfdNKYz

Drop what a stopped note's document finishes after it stops ... 52110a49cb

A seed's hash runs in real time on the thread pool. A session stopped
while it was pending carried on when it finished: it seeded its destroyed
document, stored the file's old text as its record over the next
session's, reported that text as adopted, and scheduled a write. The next
session merged the file against the stale base, and lines deleted since
came back — the property test's one-in-hundreds failure on a loaded
runner (NEC-296).

A document no longer in fileDocs now drops a late seed, stores no record
(persistRecord and beginWrite), schedules no disk write, and reports
nothing adopted (SAFE-A44). The test's IndexedDB stand-in refuses reads
and writes after destroy, as the real provider's closed database does.
Each guard was mutated alone and fails a named test.

Task: NEC-296

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YHSWLicoRvWzsYjkfdNKYz

Read nothing in for a stopped document, and check again after the last read ... d39fd9e0d2

Review: a stopped document whose read spanned the stop still read the
file in, telling the view guard, the kept-aside save and the race flag —
all kept by path — about text from a session that was gone. And the
write's liveness check came before its last awaited read, so a stop
landing in it went on to replace the file (compare-and-swap, recorded
first: nothing lost) and counted a refusal against the note.

readIn now returns for a stopped document; writeToDisk checks again
after the read and counts no refusal then. Each mutated alone fails a
named test. The docs say only what the evidence shows.

Task: NEC-296

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YHSWLicoRvWzsYjkfdNKYz

Record the soak after the fix, and what it does and does not show ... 3388531c7b

Task: NEC-296

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YHSWLicoRvWzsYjkfdNKYz

cruelacid force-pushed worktree-nec-296-a-word-deleted-from-a-note-came-back-in from 3388531c7b to 4213bd84f8 2026-10-17 02:04:43 +01:00 Compare

cruelacid force-pushed worktree-nec-296-a-word-deleted-from-a-note-came-back-in from 4213bd84f8 to 5058688cbb 2026-10-17 07:44:45 +01:00 Compare

nectenda-agent merged commit 5058688cbb into main 2026-10-17 10:40:42 +01:00

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels   

area:docs

docs/

  

area:identity

packages/identity

  

area:ops

deploy/, CI, hosting

  

area:plugin

packages/plugin

  

area:server

packages/server

  

channel:community

Obsidian forum, Obsidian Discord, Reddit

  

channel:direct

Cold outreach to named firms

  

channel:owned

nectenda.com pages, the feedback board

  

channel:press

Hacker News, newsletters, outlets, YouTube creators

  

channel:social

Bluesky, Mastodon, X, YouTube accounts

  

e2ee-constrained

End-to-end encryption changes the shape of the answer

  

gate:at-ga

Held until Gates A-C close and terms/privacy leave DRAFT

  

gate:pre-ga

Safe to do before the launch gates close

  

marketing

Reach and positioning; work published off nectenda.com

  

parity

Relay feature-parity programme

  

relay:absent

4 - Not supported, not requested, not planned at Relay

  

relay:planned

2 - On Relay's roadmap (Active/Planned) or has an open PR

  

relay:requested

3 - Asked for by Relay users or under 'Considering'; not committed

  

relay:supported

1 - Relay ships this today

  

risk:additive

New message type, table or column; backward compatible

  

risk:contract

Changes something existing accounts or stored data depend on

  

risk:none

Client-only; the server is untouched

  

usability

How the product is understood and learned, not what it does

No labels

area:docs

area:identity

area:ops

area:plugin

area:server

channel:community

channel:direct

channel:owned

channel:press

channel:social

e2ee-constrained

gate:at-ga

gate:pre-ga

marketing

parity

relay:absent

relay:planned

relay:requested

relay:supported

risk:additive

risk:contract

risk:none

usability

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

2 participants

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

Nectenda/nectenda!302

No description provided.