The UPDATE and CREATE all-day branches each called getHouseholdTimezone(db)
independently, so a drain processing both an all-day create row and an
all-day update row issued two identical app_config SELECTs. Add a lazy
per-cycle TimezoneResolver (mirroring the existing clientCache thread-through)
created in runOutboxDrain and passed into dispatchRow. The read stays lazy —
cycles with no all-day work never touch the DB — but is shared across all
all-day rows in a cycle. Behavior unchanged.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- reminderScheduler.ts: add import { getHouseholdTimezone } from '../lib/householdTimezone.js'
and replace bare process.env.TZ ?? Intl... at line 247 with await getHouseholdTimezone(db)
- outboxWorker.ts: add same import and replace BOTH bare tz lookups at the update-branch
(~line 501) and create-branch (~line 607) with await getHouseholdTimezone(db)
- D-05 satisfied: all three all-day sites now read from the single stored accessor
- D-06 satisfied: getHouseholdTimezone falls back to process.env.TZ → Intl when unset;
existing process.env.TZ-pinned tests pass unchanged
- D-07 satisfied: eventDateTime.ts and hydrateEvents.ts are not modified
- outboxWorker.test.ts: update wireMockChain() to handle app_config table with where().limit()
chain returning empty rows (D-06 fallback), so existing CAL-13 all-day test stays green
- reminderScheduler.test.ts: update mockTwoQueries to mock the new third db.select() call
(getHouseholdTimezone) returning no row (D-06 fallback), keeping all 37 existing tests green
- All 76 broker tests pass; tsc --noEmit clean
- eventFieldsSchema (events.ts): rejects reminderLeadMinutes > 10080 with 400
- outboxPayloadSchema (outboxWorker.ts): hard-fails row when value exceeds 1-week cap
- 10080 = 1 week in minutes; matches UI select maximum
classifyValarms: check sign of dur.toSeconds() before preset lookup.
Positive value = alarm fires after event (RFC 5545 TRIGGER:+PT15M or
TRIGGER;RELATED=END:PTNm) → return {kind:'custom'} for preserve path.
Compute leadMinutes as -seconds/60 (was Math.abs) for negative triggers.
Prevents alarm direction inversion: +PT15M was being stored as 15-min-before
lead and re-fired at dtstartUtc-15min — the opposite of the original intent.
- humanizeLeadMinutes: add isAllDay=false param; all-day branch returns
"Today" (lead=0), "Tomorrow" (1440), "In 1 week" (10080), "In N days" (other)
- byKey map: store isAllDay flag (false for timed, true for all-day)
- dispatch loop: pass event.isAllDay to humanizeLeadMinutes
All-day same-day reminder push now reads "Today" instead of "Starts in 0 min".
Timed event wording unchanged (isAllDay defaults to false).
- expand.ts: add reminderIsCustom:boolean to CalendarOccurrence interface;
derived from classifyValarms kind==='custom'; propagated to both
non-recurring and recurring occurrence branches
- client.ts: mirror reminderIsCustom on CalendarOccurrence (atomic mirror)
- EventForm.tsx: extend deriveReminderValue to accept isCustom flag;
returns '__custom__' when true, making the existing D-08 preserve branch
live — editing a custom-alarm event now omits reminderLeadMinutes from
the payload so outboxWorker extractValarms keeps the original VALARM
- Fix existing test fixtures (EventForm.test.tsx, EventDetailPopover.test.tsx)
to include reminderIsCustom:false on all CalendarOccurrence literals
Fixes CAL-14 Pitfall 1: Apple Calendar absolute DATE-TIME / multi-VALARM
alarms no longer silently stripped on any edit round-trip from the PWA.
- Add all-day 9 AM tests: 0-lead fires at EDT alert UTC, not midnight
- Add 1440-lead (day-before) and 10080-lead (7-day-before) tests
- Add all-day dedup test: same uid:dtstartMs fires once across ticks
- Fix all-day prune bug: store start-of-next-day as pruneMs instead of
UTC midnight (which was always <= now by fire time, causing immediate prune)
- Separate dtstartMs (dedup key component) from pruneMs (map cleanup value)
- 28/28 tests GREEN; full API suite 314/314; tsc --noEmit clean
- Add reminderLeadMinutes: number | null to CalendarOccurrence interface (D-06)
- Import classifyValarms in expand.ts; derive series-level value once per VEVENT
- Add reminderLeadMinutes to both non-recurring and recurring occurrence construction
- Add reminderLeadMinutes to GET /api/events select for edit-mode pre-population
- Replace fixed 16-min window with per-event variable-lead fire-time check
- Two separate DB queries: timed (allDay=false) + all-day (allDay=true)
- Remove eq(calendars.isShared, true) — personal events now dispatch (NOTIF-05)
- Remove eq(calendarEvents.allDay, false) — all-day handled in separate query
- Add reminder_lead_minutes IS NOT NULL WHERE predicate (NOTIF-05)
- Skip timed events with reminderLeadMinutes===0 in JS (D-06: 0 on timed = None)
- Change dedup key from bare uid to uid:dtstartMs compound key (NOTIF-06)
- Update prune loop to use compound key
- Import computeAlertInstantUtc from vevent.js (Plan 11-01, wave 2 dep)
- Add humanizeLeadMinutes export (Task 2 body formatter, used in dispatch)
- Update test helper mockTwoQueries() to handle two-query dispatch pattern
- All 14 tests GREEN; tsc --noEmit clean; setInterval retained, no node-cron
- Import classifyValarms from vevent.ts
- Derive reminderLeadMinutesValue: preset/offlist → leadMinutes; custom/none → null
- Add reminderLeadMinutes to .values() and .onDuplicateKeyUpdate({ set: {} })
- Scheduler now has ground truth for native-client VALARMs (T-11-07 mitigated)
- eslint.config.js: disable React Compiler rules (v7 flat.recommended enables
them; codebase does not use the Compiler); add e2e/ to disableTypeChecked
block; promote exhaustive-deps to error
- API broker: remove redundant as-casts (outboxWorker, poller, reminderScheduler,
expand, sync, vevent, spike); add targeted ical.js no-unsafe-assignment/argument
disables with justifying comments inside try blocks
- API routes/sse.ts: fix no-misused-promises on async writeSSE callback with
void+IIFE+catch pattern
- API routes/lists.ts: let → const for updateValues
- API tests: remove unused imports (beforeEach, eq, vi); rename unused vars
with _ prefix; remove unused lastActiveId assignment
- PWA components: void navigate() and void queryClient.invalidateQueries() on
all fire-and-forget call sites; fix CalendarShell explicit-type-casts;
Couldn't → HTML entity
- PWA test files: as unknown as Response for partial mock objects; string | null
type annotation on mockLastSyncedUid; remove async from test callbacks without
await; act(() => {}) not await act(async () => {}) for sync ops
- sw.ts: restructure Notification.data?.url access as let+if so disable
comments land on the exact violation lines; void self.skipWaiting()
- Replace [now+14min, now+16min] window with (now, now+16min] catch-up
- Replace minuteBucket-keyed Set with uid-keyed Map for exactly-once dedup
- Lead-accurate body: 'Starts in N min' (Math.max(1, round(lead/60000)))
- CR-01 pruning: drop entries whose dtstart <= now (event started)
- WR-01 preserved: mark uid sent after all dispatches complete
- Drop gte import; add gt import from drizzle-orm
- Add hasRrule: boolean to CalendarOccurrence interface
- Capture isRecurring = event.isRecurring() once before the branch
- Set hasRrule: isRecurring in non-recurring push (always false)
- Set hasRrule: isRecurring in recurring push (always true)
- All 10 expand.test.ts tests pass (RED→GREEN)
Pre-capture all currently-cached rows into pendingDeleteRows before the whole-cache
db.delete() when seenUids.length === 0. The existing >0 branch behavior is unchanged.
Adds a regression test verifying onChanges receives one delete change per cached row
on a full-calendar clear.
- Create eventChangeDispatcher.ts: dispatchEventChange + isMeaningfulChange
- D-04: description-only edits are silent; meaningful fields = title/dtstartUtc/dtstartDate/allDay/location
- D-03: actor excluded via ne() + application-level filter; all subs filtered by userId != actorUserId
- D-13: reads only push_subscriptions from MariaDB — no tsdav/Fastmail I/O
- syncCalendar: add optional onChanges callback; populate title from VEVENT SUMMARY on every upsert
- syncCalendar: pre-upsert SELECT to detect add vs update; track changedFields; prune emits deletes
- poller: pass onChanges with actor=cred.userId (external changes from other member)
- outboxWorker.triggerTargetedResync: pass onChanges with actor=userId (this-member writes)
- All 4 eventChangeDispatcher tests + 14 sync tests GREEN
- Create apps/api/src/broker/reminderScheduler.ts:
- runReminderCheck(now): queries isShared=true + allDay=false events
with dtstartUtc in [now+14min, now+16min] via calendarEvents→calendars→
pushSubscriptions cross-join (2 innerJoins; fans out to all subscribers)
- In-memory sentReminders Set keyed uid:minuteBucket prevents double-fire
at window boundary (D-06, T-05-18)
- Per-event and per-subscription try/catch for error isolation (T-05-19)
- Null title fallback (event.title ?? uid) — handles rows before Plan 05-07
- Empty shared-calendar set produces zero sends and no crash (D-16)
- startReminderScheduler(): node-cron 1-min schedule wrapping runReminderCheck
- Wire startReminderScheduler() into index.ts isMainModule() guard after
startOutboxWorker() and VAPID setVapidDetails (NOTIF-01)
- Tests: 3/3 GREEN (all-day excluded, non-shared excluded, dedup)
Two write-path cache bugs surfaced during Gate 2 live testing:
P1 (delete didn't work / ghost event): syncCalendar only UPSERTED events
present on Fastmail and never removed cache rows for events that disappeared.
A successful CalDAV delete left the row in calendar_events forever, so
GET /api/events kept returning it and the UI showed a ghost that 'wouldn't
delete' (even after refresh). Add a prune step: delete calendar_events rows for
this calendar whose uid is absent from the server response (scoped to cal.id so
it never touches another calendar or the other member's rows — BUG B). Empty
server result prunes the whole calendar's cache.
P2 (edit needed a manual refresh): the outbox worker marked a row 'done' BEFORE
triggerTargetedResync refreshed the cache. The PWA's SyncStateToast invalidates
['events'] the instant sync-status flips to 'done', so it refetched stale cache.
Re-sync first, then mark done — 'done' now guarantees the cache reflects the write.
Tests: +2 prune regressions (present-subset prune, empty-server prune-all).
BUG A — timed events written 4h off: EventForm sent a naive local wall-clock
string with no offset; the UTC API container parsed it via new Date() as UTC, so
09:00 America/Toronto serialized to DTSTART:...090000Z. Fix: new
apps/pwa/src/lib/eventDateTime.ts serializes timed events to an unambiguous UTC
instant in the browser (where the operator's zone is known); all-day stays a DATE
string. No backend change.
BUG B — created events attached to the wrong user's calendar + duplicate calendar
rows per poll: calendars had no unique key on url, and poller/sync matched
calendars by url alone — so under the shared single Fastmail account (D-16) one
member's collection resolved to the other member's row. Fix: composite
unique(user_id, url); scope poller lookup + sync select to (userId, url); hand
migration 0001 (dedup + add key), applied to the live DB.
Regression tests fail against the buggy url-only predicate. API 98/98, PWA 140/140,
tsc clean both packages.
- In update dispatch, SELECT etag FROM calendar_events WHERE uid = row.uid before PUT
- Use fresh etag as If-Match instead of stale enqueue-time row.etag when available
- Fall back to row.etag when calendarEvents has no matching row
- D-08 conflict detection intact: genuine external changes update calendarEvents.etag
differently from any pending row, so they still 412 correctly
- outboxWorker: remove empty-credential fallback; let loadClientForUser throw on error (CR-03)
- outboxWorker: fix backoff index from nextAttemptCount to row.attemptCount so first retry waits 15s not 60s (WR-01)
- events.ts: replace bare crypto.randomUUID() with import { randomUUID } from 'node:crypto' on all three handlers (WR-08)
