NEC-309: A Kanban card deleted by another writer comes back when the board view saves an edit begun before the delete #321

Merged
nectenda-agent merged 12 commits from worktree-nec-309-a-kanban-card-deleted-by-another-writer into main 2026-10-26 14:06:29 +00:00
Collaborator

Task: NEC-309 — https://projectron.nerchure.com/tasks/328

A Kanban board holding an edit it had not saved lost another writer's change through Obsidian 1.13.7's own three-way merge (diff_main with a zero deadline), and our read-in spread the loss: a deleted card came back, an added one was deleted with no copy. The board's next save is now read against what it had loaded (SAFE-A50), with a copy and notice where a line would be left out. Rides along: SAFE-A39 timed from another writer's change (NEC-338); an offline edit merged against the bound editor's own text as sync stopped (NEC-341); an image card's fitted size no longer undoes a concurrent delete (NEC-343); an unbound editor's edits the document already holds are not replayed (NEC-344). Each found by a hardness run, reproduced on main first, and mutation-checked; two fresh-context reviews, findings fixed. Spec: docs/changes/NEC-309-kanban-delete-by-another-writer/spec.md.

Spec: docs/changes/NEC-309-kanban-delete-by-another-writer/spec.md

Also fixes NEC-338: An edit begun just after another writer's change to a Kanban board is not marked as raced
Also fixes NEC-341: A note's deletions made offline come back at the reconnect's merge
Also fixes NEC-343: A canvas card deleted in one vault comes back in both
Also fixes NEC-344: A token is dropped from a note renamed offline, kept only in a conflicted copy

Changelog

A Kanban card another plugin or tool deleted or added while you were editing the board no longer comes back or disappears in every vault.

Task: NEC-309 — https://projectron.nerchure.com/tasks/328 A Kanban board holding an edit it had not saved lost another writer's change through Obsidian 1.13.7's own three-way merge (diff_main with a zero deadline), and our read-in spread the loss: a deleted card came back, an added one was deleted with no copy. The board's next save is now read against what it had loaded (SAFE-A50), with a copy and notice where a line would be left out. Rides along: SAFE-A39 timed from another writer's change (NEC-338); an offline edit merged against the bound editor's own text as sync stopped (NEC-341); an image card's fitted size no longer undoes a concurrent delete (NEC-343); an unbound editor's edits the document already holds are not replayed (NEC-344). Each found by a hardness run, reproduced on main first, and mutation-checked; two fresh-context reviews, findings fixed. Spec: docs/changes/NEC-309-kanban-delete-by-another-writer/spec.md. Spec: `docs/changes/NEC-309-kanban-delete-by-another-writer/spec.md` Also fixes NEC-338: An edit begun just after another writer's change to a Kanban board is not marked as raced Also fixes NEC-341: A note's deletions made offline come back at the reconnect's merge Also fixes NEC-343: A canvas card deleted in one vault comes back in both Also fixes NEC-344: A token is dropped from a note renamed offline, kept only in a conflicted copy ## Changelog A Kanban card another plugin or tool deleted or added while you were editing the board no longer comes back or disappears in every vault.
Obsidian 1.13.7 merges a change to the file into a view holding unsaved
text three ways, with diff_main(..., true, 0): a diff given no time to run.
Where the view's text differs from what it loaded in more than a couple of
places, the merge comes out as the view's own text, and the change is gone
from the view. Its save, read in against the change, then undid it in
every vault: a card another writer deleted came back (Markdown hardness
run, seed 281144514, round 159), and one it added was deleted with no copy.
Replayed offline on that run's own texts, the merge gave exactly the save.

The guard now notes each open view inside the change's modify event, before
Obsidian's reload has read the file: what it would save, and what it last
loaded or saved. A view that held an edit it had not saved has its next save
read against what it had loaded, so only its own edits come in. Where that
leaves out a line of the save, the save is kept first and the user told
(SAFE-A50). Only for another plugin's view of a note.

Task: NEC-309
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012fdZKBVPZM2KxUuJMLmneg
A board holding an edit it had not saved, saving after another writer's
change, has its save kept before it is merged wherever the merge leaves out a
line of it: in seed 281144514's round 159, the card the other writer deleted.
Without this the checker called that copy one with no reason for it. Read
from the plugin's own log line, as SAFE-A39's keeps are, and it excuses
nothing missing from a note.

Task: NEC-309
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012fdZKBVPZM2KxUuJMLmneg
SAFE-A39 marks an edit begun in another plugin's view of a note within a
second of our write: Kanban opens a card's editor on the board it showed
before the reload and commits the edit against that layout, over another
card's line (NEC-280). Obsidian reloads the view with another writer's
change the same way, and nothing marked an edit begun then.

The mark is now also timed from another writer's change, from its modify
event, and confirmed when the change is read in: a change that was the
view's own save reloads nothing and marks nothing. An edit begun before the
read-in finishes is noted and marked then. The copy and notice are SAFE-A39's
as before; the notice says "a change" rather than "a remote change".

Task: NEC-309
Also-fixes: NEC-338
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012fdZKBVPZM2KxUuJMLmneg
A board that saved as it closed, or that Kanban edited again before its
save was read in, was gone or held something else by the read-in, so its
save lost the base SAFE-A50 gives it and was read against the other
writer's change after all: an added card deleted with no copy, the fault
this card is about. The save's modify event already noted what each view
held; the base is now noted beside it and matched from there.

Also from review: whether a save is one SAFE-A50 copies is decided with its
base, before anything awaits, and a ride-along mark noted before a slow
read-in is confirmed however long the read-in took.

Task: NEC-309
Also-fixes: NEC-338
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012fdZKBVPZM2KxUuJMLmneg
A collaborator's change reached a bound editor, sync stopped, and the
editor saved what it held as it let go; offline, the user edited from
there. The next connect merged the file against the text disk and
document had last agreed on, from before the collaborator's change. Both
sides had then changed the same lines, both versions were kept, and a word
deleted offline came back in both vaults (Markdown hardness run, seed
407628096, round 39, the same on main).

As sync tears a note down with an editor bound, the editor's own text is now
recorded as a let-go text: what it saves as it lets go, never the
document's, which can hold a change the editor did not show. The next
connect merges against it where the file shows its change, and against the
agreed text otherwise, as when the editor's save never landed.

Task: NEC-309
Also-fixes: NEC-341
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012fdZKBVPZM2KxUuJMLmneg
Obsidian fits an image card to its image's shape as the image loads,
with no gesture and no save (updateAspectRatio). A live view's next
capture, for any other edit, carried that size into the document as this
vault's edit. Concurrent with a delete of the card elsewhere, the delete's
snapshot no longer matched and the card came back in both vaults
(canvas hardness run, seed 1717489251, round 72, the same on main).

A width- or height-only change to a file card, with no pointer gesture
in the view since its last capture, is now taken as the fit: remembered
for the card, kept in the view, never carried. The view's own save that
differs from what it held only by fits is still its own. A size chosen
by a gesture is carried as any edit.

Task: NEC-309
Also-fixes: NEC-343
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012fdZKBVPZM2KxUuJMLmneg
Keeping the fit out of the document, as the commit before did, left two
vaults' files disagreeing by it for the rest of the session: the canvas
hardness run found that in its first hour (seed 2118765156, round 4).

The fit is carried again, so every vault shows one board, but the view
that carried it remembers it: a width- or height-only change to a file
card with no pointer down on that card since the last capture, and no
drag of it. A tombstone arriving that the record misses only by that fit
is made good by a tombstone of this vault's holding the record as it is,
before the draw that would bring the card back. A size the user chose by
a gesture on the card keeps it, as any edit does (SAFE-A18).

The review found that any pointer down had counted: a click to select
the card about to be recoloured made another card's fit look like a
resize. Only the card the pointer went down on counts now.

Task: NEC-309
Also-fixes: NEC-343
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012fdZKBVPZM2KxUuJMLmneg
Both from the second review. A change that removed a blank line, met by a
file a tool had removed another blank line from, read as shown, and a
paragraph the change added was taken as deleted: blank lines count for
nothing in showsChange now. And the guard's change record noted a view
once per record, so a second save within a second of the first, before
the first was read in, had no state of its own, and a board closed by
then no base for it: one state per text the view held.

Task: NEC-309
Also-fixes: NEC-341
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012fdZKBVPZM2KxUuJMLmneg
Task: NEC-309
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012fdZKBVPZM2KxUuJMLmneg
B, offline, deleted three tokens in an editor that had let go of its
note, and renamed the note. The editor's save landed with sync stopped,
so nothing cut the edit from its log; the renamed note's document was
seeded from that save, and the bind replayed the log onto it. Placed
through a character alignment, 7 of the 36 deleted characters matched
the look-alike start of the next token, and a collaborator's token was
cut in both vaults (Markdown hardness run, seed 1209056253, round 57,
the same on main).

A log whose edits make of its base exactly the text the document holds
is now not replayed: the document has them (SAFE-B5).

Task: NEC-309
Also-fixes: NEC-344
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012fdZKBVPZM2KxUuJMLmneg
Record NEC-309's hardness runs in its spec
Some checks failed
CI / promote (pull_request) Has been cancelled
CI / e2e (pull_request) Has been cancelled
CI / build (pull_request) Has been cancelled
Release note / release-note (pull_request) Successful in 19s
8181e7cd95
Task: NEC-309
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012fdZKBVPZM2KxUuJMLmneg
Add NEC-309's release notes, one line per change
All checks were successful
Release note / release-note (pull_request) Successful in 16s
CI / build (pull_request) Successful in 5m42s
CI / e2e (pull_request) Successful in 14m32s
CI / promote (pull_request) Has been skipped
CI / build (push) Successful in 6m33s
CI / e2e (push) Successful in 14m34s
CI / promote (push) Successful in 31s
0aa18e0cd8
Task: NEC-309
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012fdZKBVPZM2KxUuJMLmneg
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
2 participants
Notifications
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!321
No description provided.