8000 fix: follow-ups from the v0.48.0 review by kvaps · Pull Request #306 · aenix-io/aeman · GitHub
[go: up one dir, main page]

Skip to content

fix: follow-ups from the v0.48.0 review - #306

Merged
kvaps merged 17 commits into
mainfrom
fix/v0481-followups
Nov 7, 2026
Merged

kvaps merged 17 commits into
mainfrom
fix/v0481-followups

Conversation

@kvaps
@kvaps kvaps commented Nov 7, 2026
Copy link
Copy Markdown
Member

Follow-ups from the final review of v0.48.0.

  • Watch frames counted a plan item's tasks over the whole board, including domains the watcher can't read. They now count what the watcher may read, the same as the listing.
  • A listing that arrived after a delete could bring the deleted card back. Listings and the drawer's cache now skip cards this tab just deleted.
  • A delete inside a request now rides that request's write queue, so unassigning a plan is one commit per repository and a delete never overtakes earlier writes. If the queued delete fails, the card comes back at once for everybody and the tab forgets its delete marks.
  • Delete commits are named for the card (delete «…»); unassign commits for the action (take 3 tasks out of work under «…»).
  • status.subtask tells a client whether a card rides a parent task; a review card with its own parent no longer drops off Team and Me once taken. A review of a review inherits its priority through the chain.
  • Fixed: the off-board × on a card with a subtask failed with "card not found".
  • Smaller: a refused priority rolls back only the latest press; a valid priority write leaves one priority: key; Unassign with nothing in work says so instead of asking about zero tasks; the drawer tree picks up children created elsewhere; Back skips a deleted card; one backlog-order fixture shared by the server and the client; docs for the backlog order.

kvaps added 17 commits November 7, 2026 22:23
An invalid priority value rides through a card file as an unknown key.
Setting a valid one then wrote both, a duplicate YAML key that broke
every later parse of the card. The unknown value now gives way to the
valid one.

Signed-off-by: Andrei Kvapi
8000
l <kvapss@gmail.com>
The view parameter, the list_cards tool and the backlog listing still
said a backlog is read by hand, then oldest first. It is read by
effective priority, each priority led by the reviews nobody has taken,
then in the Roadmap's order.

Signed-off-by: Andrei Kvapil <kvapss@gmail.com>
Tombstones guarded only the watch: a list request sent before a delete
and answered after its DELETED frame still carried the card, and
merging it into the board, or caching it for the drawer's tree,
brought the card back until the next reload. Listings are now filtered
through the same marks.

Signed-off-by: Andrei Kvapil <kvapss@gmail.com>
A priority press that the server refused restored the state before it
even when a newer press had been shown since, so an older P1 failing
wiped the optimistic P2. The rollback now happens only while the
refused press is still the latest on its card.

Signed-off-by: Andrei Kvapil <kvapss@gmail.com>
When the server's dry run found nothing in work under the row any
more, the Roadmap still opened "Take 0 tasks out of work?". The
confirm now opens only when there is a task to take. Also drop a stale
project-card comment and the identity helper the preview replaces.

Signed-off-by: Andrei Kvapil <kvapss@gmail.com>
A watch frame's status.hierarchy was counted over the whole board,
while the listing beside it counts over the watcher's visible board:
an epic over tasks in a closed repository read one way in the list
and another on the stream, which told the watcher how much work sat
where they cannot look. The frame now counts only the planning cards
and tasks its watcher can read.

Signed-off-by: Andrei Kvapil <kvapss@gmail.com>
A review card inherits through its task's ancestors, which are on no
crumb of its own, so a watcher who could read the task but not its
parent task was sent that parent's priority and title in the review's
frame. The task's chain now counts toward the frame's hidden check, as
the listing on the visible board already does.

Signed-off-by: Andrei Kvapil <kvapss@gmail.com>
A card made in another tab under the card the drawer shows now joins
its "Under it" tree from the ADDED frame, when its parent chain
reaches that card. A card deleted outright also leaves the drawer's
history, so Back no longer opens a card that is gone.

Signed-off-by: Andrei Kvapil <kvapss@gmail.com>
A card deleted during a request — a review withdrawn when UNASSIGNED
parks its task — was deleted synchronously: a commit of its own, with
no action id, landing ahead of the request's other writes. It now
rides the request's queue, so the whole request is one commit.

Signed-off-by: Andrei Kvapil <kvapss@gmail.com>
board.BacklogOrder and the client's backlogList now read the same
fixture, with each card's wire status checked against CardResource.
Two cases disagreed: a review with a stray parent of its own carried
that parent on its lineage, and a review of a review was placed by
itself on the server and by the review it reviews on the client. A
review card's lineage is now the reviewed card's planning chain alone,
and a review of a review stands by the review it reviews on both
sides.

Signed-off-by: Andrei Kvapil <kvapss@gmail.com>
UNASSIGNED under a plan item is one commit over many tasks, and it took
its subject from whichever task was parked first. The request now names
its commit for what it did: "take 11 tasks out of work under «ТС-2»",
or "take «task» out of work" on a single task.

Signed-off-by: Andrei Kvapil <kvapss@gmail.com>
A SyncError may be one of this tab's queued deletes failing, and the
server sends the card again. The tombstone kept turning it away and
kept it out of listings for a minute. Every mark is dropped on a
SyncError, since the tab cannot tell which delete failed.

Signed-off-by: Andrei Kvapil <kvapss@gmail.com>
A review card's lineage is the reviewed task's chain, so the client
read a stray parent of its own as a task parent and took the review
for a subtask: once taken, the Team and Me boards stopped drawing it
while the server still listed it. The server now says so in
status.subtask, the client reads it for review cards, and the shared
backlog fixture holds both sides to one answer.

Signed-off-by: Andrei Kvapil <kvapss@gmail.com>
A delete inside a request now rides the request's queue and marks the
card gone at once. When the write then failed, only the first op's
recent-write guard was cleared and the recent-delete guard stayed, so
every reload kept dropping the card for the whole grace window; and the
reload a failure triggers was not told to watchers, since the failed
write had marked the cache unloaded. The failed group now clears both
guards for every card it took, and the reload is sent as a diff, so the
card comes back as ADDED on every open board.

Signed-off-by: Andrei Kvapil <kvapss@gmail.com>
The backlog promised that a review of a review stands with the review
it reviews, but its effective priority stopped at a review and it fell
into the no-priority group. The priority now follows reviews through
to the task on both sides, and a cycle of reviews counts under none.

Signed-off-by: Andrei Kvapil <kvapss@gmail.com>
When the server's preview found nothing in work under the row, the
press did nothing at all and read as a broken button. It now says
"Nothing in work here", the disabled item's own wording.

Signed-off-by: Andrei Kvapil <kvapss@gmail.com>
A delete frees or takes along subtasks, or puts a reviewed task back
In progress, before the card itself goes, and the request's commit was
named for the first of those writes. The delete and the board's x now
name it for the card. Taking a card off with its subtasks also read the
board again before freeing what was left: freed from the stale board,
it rewrote a subtask it had just deleted, and the store refused the
whole request. The matrix records the queued delete as G88, and W6
follows.

Signed-off-by: Andrei Kvapil <kvapss@gmail.com>
@kvaps
kvaps merged commit d2cea70 into main Nov 7, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant

0