Deck's Entity getters are magic (__call), so method_exists() never saw
getDeletedAt() or getArchived(): trashed cards were moved, trashed boards
were neither dropped from "my boards" nor detected by findStackIds(), and
every board counted as not archived. DeckIntegrationService now checks the
properties with property_exists() via isTrashed()/isArchived().
Workflows on archived boards stay enabled but are skipped by the runner
(isBoardArchived(), read through BoardMapper) until the board is
unarchived, and the settings list marks them. Archived boards are no
longer offered in the board dropdown at all; editing a workflow on one
clears board and stacks and says why.
Every enabled workflow is an edge "source stack -> target stack";
LoopDetector finds circles in that graph, a workflow whose source is its
own target being the shortest. Without a check, two workflows A -> B and
B -> A move the same cards back and forth on every run, even within a
single run, with a notice for each move. The trigger date is ignored on
purpose: by start date one way and by due date the other, a card with
both dates passed still circles. Chains stay allowed.
The controller rejects saving an enabled workflow that would close a loop
and names the workflows involved. Workflows stored before this check are
skipped by the runner (logged, not disabled) and flagged in the settings
list via a new `loop` field from index().
Each workflow now has a trigger date (due or start): the runner compares
either the card's due date or its start date against now, read straight
from Deck's Card entity. Mails and Nextcloud notifications say whether
the card was overdue or reached its start date. Cards without the chosen
date stay put.
Deck only has a start date since 1.18 and Entity getters are magic, so
DeckIntegrationService::supportsStartdate() checks the property; without
it the form rejects start-date workflows and the runner skips them
without disabling them.
WorkflowRunner::isOverdue() becomes hasPassed(), as it now covers both
dates. Existing workflows keep the due date (migration Version1005).
Bumps to 35.4.0 so the migration runs.
Each workflow now has a notify channel (mail, notification or both) next
to its recipient mode. Notifications go out per card through
OCP\Notification\IManager, so they show up in the bell and as push on the
mobile and desktop clients. Without the notifications app nothing is sent.
The one-off "workflow disabled" notice is sent as a notification as well.
Existing workflows keep mail (migration Version1004). Bumps to 35.3.0 so
the migration runs.
The README now explains setup, workflows, the mail options with Deck's
real menu names, and what happens when boards or stacks disappear -
without class names. Architecture, data model, tests, CI, release and
the time_sensitive troubleshooting move to DEVELOPMENT.md, fixing the
stale CI table, PHP version and Nextcloud version on the way.
DEVELOPMENT.md is excluded from the app package.
The app is used internally only and its settings page was German
already. Mails, controller error messages (shown in the form), the
settings section name and the mail sender are now written in German
directly instead of going through IL10N without any translation files.
Nextcloud's own mail footer is requested in German as well.
Replaces the yes/no notify_email flag with notify_mode:
- off: no mail
- owner: only the board owner, as before (migrated from notify_email=1)
- board: everyone with access to the board, filtered by their own Deck
"Notifications" setting for it (board:<id>:notify-due: off /
assigned / all), mirroring Deck's NotificationHelper rules
Recipients are resolved via PermissionService::findUsers() (owner, ACL,
groups and circles expanded), AssignmentMapper::isUserAssigned() and
IConfig, none of which reads Deck's frozen $userId. If they cannot be
resolved, no mail is sent rather than one to the wrong people.
Moved cards are collected over the whole run and each recipient gets a
single digest mail afterwards, in their own language.
Bumps to 35.2.0 so the migrations run.
Workflows now move every overdue card of their source stack. A new
migration drops filter_user_ids and filter_label_ids; workflows that
had a filter keep running unfiltered. The version goes to 35.1.0 so
the migration runs on update.
Removed along with the filters: the labels and participants endpoints
and their Deck lookups, the dead-filter warnings and markers, and
CardService::enrichCards() in the card read. enrichCards() only added
labels, assignees and comment counts, and ran a permission check
against Deck's frozen uid.
Nextcloud Hub Summer 26 (server 35.x) is declared as the sole
supported major on this branch, per the existing versioning scheme
(leading version number tracks the declared NC major). main keeps
tracking Nextcloud 34 (Hub 26 Spring) at the 34.x.x line; the two are
maintained in parallel rather than one replacing the other.
revert Let each workflow watch the due date or the start date
Deck 1.18 added a start date to cards (Card::$startdate), so a workflow
no longer has to mean "overdue". Each rule now carries a date_field
('due' | 'start'), chosen in the settings form next to board and stacks.
Existing rules keep the due date: the new column's default supplies it
for every stored row, so there is no backfill step, and
Workflow::getDateFieldOrDue() covers entities that were never near the
database as well as hand-edited values. The controller is the one place
that rejects an unknown value instead of normalising it - a client
asking for a rule it would not get should hear about it.
The probe for the field is property_exists(), not method_exists():
Deck's entities declare no getter as real code - getStartdate(),
getDuedate(), even getId() all go through Entity::__call(), which
method_exists() ignores by definition. A method_exists() guard would
have been false on every Deck version and would have turned every
start-date rule into a silent no-op.
assertDeckAvailable() now also refuses Deck older than 1.18.0. That
floor cannot live in appinfo/info.xml - neither the server's nor the app
store's schema has an app-to-app dependency element - but
<nextcloud min-version="34"> already implies it, since 1.18.x is the
only Deck release line published for Nextcloud 34.
isOverdue() becomes hasPassed(): one function for both dates, since the
comparison and the argument against Deck's day-truncating
getDaysUntilDue() are identical for either.
Also carries a pending NcSelect icon-alignment fix that was already in
the working tree.
Bump to 34.1.0.
Deck 1.18 added a start date to cards (Card::$startdate), so a workflow
no longer has to mean "overdue". Each rule now carries a date_field
('due' | 'start'), chosen in the settings form next to board and stacks.
Existing rules keep the due date: the new column's default supplies it
for every stored row, so there is no backfill step, and
Workflow::getDateFieldOrDue() covers entities that were never near the
database as well as hand-edited values. The controller is the one place
that rejects an unknown value instead of normalising it - a client
asking for a rule it would not get should hear about it.
The probe for the field is property_exists(), not method_exists():
Deck's entities declare no getter as real code - getStartdate(),
getDuedate(), even getId() all go through Entity::__call(), which
method_exists() ignores by definition. A method_exists() guard would
have been false on every Deck version and would have turned every
start-date rule into a silent no-op.
assertDeckAvailable() now also refuses Deck older than 1.18.0. That
floor cannot live in appinfo/info.xml - neither the server's nor the app
store's schema has an app-to-app dependency element - but
<nextcloud min-version="34"> already implies it, since 1.18.x is the
only Deck release line published for Nextcloud 34.
isOverdue() becomes hasPassed(): one function for both dates, since the
comparison and the argument against Deck's day-truncating
getDaysUntilDue() are identical for either.
Also carries a pending NcSelect icon-alignment fix that was already in
the working tree.
Bump to 34.1.0.
A filter whose entries have all been deleted from the board was the one
failure mode of this app that was completely invisible: the workflow kept
running, kept updating last_run and matched nothing, forever. It happens
with deleted labels and just as easily when a filtered user loses board
access, because Deck's BoardService::deleteAcl() calls
assignedUsersMapper->deleteByParticipantOnBoard() and wipes that user's
card assignments.
Two low-stakes signals, no new column and no migration:
- WorkflowRunner::warnAboutDeadFilters() logs a warning on every run.
- workflowIssues in PersonalSettings.vue marks the row red.
Deliberately no mail and no `enabled = false`. One deleted label out of
three is harmless, and even a fully dead filter can be one board edit away
from being live again.
Both obey the rule the disable path already follows: only positive
knowledge. DeckIntegrationService::findFilterOptions() therefore throws
instead of degrading to [], unlike listLabels()/listParticipants(), where
an empty list only means an empty dropdown.
That also fixes an existing false positive of the same family:
loadStacksFor() stored [] when the request failed, so a single failed
fetch reported "Quell-Stapel und Ziel-Stapel nicht mehr vorhanden" for an
untouched board. Failed requests now store null and are read as "unknown".
The comparison itself lives in WorkflowRunner::findDeadFilters(), static
and side-effect free like cardMatchesFilters(), with unit tests -- notably
that an empty filter is never dead, which would otherwise flag every
unfiltered workflow.
Separately, the moved-card mail now names the board and both stacks.
DeckIntegrationService::describeTargets() resolves the titles at most once
per workflow run, and only when a card actually moved and the workflow
wants a mail; unreadable titles degrade to #<id> rather than costing the
user their notification.
Workflows were being switched off with "its board no longer exists or is
no longer available to you" for boards that were perfectly intact.
Deck does not read the session for permissions. PermissionService -- the
class behind every check, including the ones inside CardService::reorder()
-- takes the current user as a plain `private ?string $userId`, filled
from the app container's `userId` service (ISession::get('user_id'),
registered shared). Pimple resolves that once per process and caches it,
and ServerContainer caches Deck's app container just as long. In cron the
value is null whenever a Deck-owned job ran earlier in the same pass, and
otherwise the first workflow owner touched -- never the user being
impersonated. null fails every check, so Deck answered NoPermissionException
for an untouched board and findStackIds() read that as "the board is gone".
IUserSession::setUser() never had any effect on this path.
- DeckIntegrationService::beginUserContext()/endUserContext() pin
PermissionService (mandatory), CardService, BoardService and
ActivityManager (best effort) to the workflow owner, and restore them
afterwards. If PermissionService cannot be pinned, the runner skips that
user instead of acting under someone else's permissions.
- findStackIds() now takes the uid as an argument and is assembled from
pieces that cannot answer for the wrong user: BoardMapper and StackMapper
carry no user state, and getPermissions() is handed the uid explicitly.
It no longer goes through StackService::findAll().
- NoPermissionException is no longer treated as "board missing". Unknown
failures throw, which leaves the workflow enabled.
This also fixes the second half of the same defect: card moves silently
failed for every user except the first one processed in a cron pass.
Separately, RunWorkflowsJob is now a plain Job instead of a TimedJob and
runs on every cron pass. The workflow_deck_automation.interval config key
is gone. Time sensitivity is no longer merely declared but structurally
unreachable: JobList::add() leaves the column at its TIME_SENSITIVE
default, and the ratchet in setLastRun() only fires for TimedJob.
Deleting a board or stack left the workflow row pointing at nothing: Deck
keeps the orphaned cards readable until its DeleteCron purges them, so the
run kept failing on the move every few minutes, silently and forever.
- The controller now rejects boards and stacks that do not exist (or are not
the user's) on create and update, so no new broken row can be stored.
- The settings list marks affected workflows in red instead of showing a row
that looks healthy.
- The background job disables such a workflow and mails its owner once. The
mail ignores the notifyEmail flag: that one is about moved cards, this is a
notice that the automation stopped. It stays a one-off because a disabled
workflow is no longer picked up.
The check deliberately goes through StackService::findAll() rather than
BoardService::getUserBoards(): Deck injects the current user into BoardService
as a string frozen at construction, so in the job -- one process, many users,
a cached BoardService -- it would answer for the wrong user or for none, and
every workflow on the instance would have been disabled. findStackIds()
returns null only for a genuinely missing or forbidden board and throws for
anything else, so a Deck outage skips the workflow instead of killing it.
Shipping the TIME_SENSITIVE change is not enough for instances that already
ran the job: getNext() filters on the time_sensitive column of oc_jobs, and
JobList::setLastRun() only ever sets it to TIME_INSENSITIVE -- there is no
else branch, and JobList::add() writes the column only when inserting, so
re-registering an existing job on app update or app:enable leaves the stamp
in place.
Document the one-off UPDATE (and the occ-only alternative) in the README
troubleshooting section and in CLAUDE.md.
The job was registered but never executed: `background-job:list` kept
showing last_run = 1970-01-01. It was marked TIME_INSENSITIVE, and
OC\Core\Service\CronService::runCli() reads `maintenance_window_start` and
calls jobList->getNext($onlyTimeSensitive = true) whenever the current UTC
hour is outside [start, start+4] -- so with a maintenance window configured,
time-insensitive jobs are skipped for the other 20 hours of the day.
A job whose entire purpose is a 5-minute (now configurable down to 60s)
reaction time is time-sensitive by definition.
Also fix the occ invocation in README.md and CLAUDE.md: background-job:worker
takes job classes, not an app id, so the documented
`background-job:worker workflow_deck_automation` matched nothing.
- Boards: BoardService::getUserBoards() defaults to $includeArchived = true,
and that flag also gates the `deleted_at = 0` condition, so archived and
trashed boards showed up in the dropdown. Request the filtered query
instead, with a fallback to the no-arg call if the signature ever changes.
Stacks need nothing: StackMapper::findAll() always filters deleted_at.
- Each stack dropdown now hides whatever the other one holds, so source and
target can no longer be set to the same stack. The save-time check stays
as the backstop for rows stored before this rule.
- RunWorkflowsJob reads its interval from config.php
('workflow_deck_automation.interval', seconds, default 300, clamped to a
60s minimum). TimedJob re-reads the interval on every cron pass, so a
changed value takes effect without any occ command.
build-release.yml and build-main.yml ran identical steps (checkout, node,
npm install, npm run build, make appstore) and differed only in where the
package landed. Fold the v* tag trigger into build-main.yml and branch on
gitea.ref in the publishing step: main keeps the rolling latest-main
pre-release, tags get a normal release on the tag that was just pushed.
Tag deletion is now guarded - only latest-main is recreated, since deleting
a version tag would destroy the ref that triggered the run. A stale release
is still removed in both cases so moved tags republish cleanly.
Side effects: version builds are published as durable release assets with a
stable URL instead of expiring workflow artifacts, which drops the last
actions/upload-artifact usage and with it the @v3 pin forced by the GHES
restriction; and the setup-php step goes away, since make appstore only
shells out to mkdir/tar/rm.
Ports the standalone due-date cron script into a proper Nextcloud 34
app with a personal-settings UI, per-user workflow configuration
(source/target stack, assigned-user and label filters, email
notification), a TimedJob background runner, and Gitea CI pipelines.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Patrick Niebeling
gnilebein
