Commit Graph
143 Commits
Author SHA1 Message Date
makiolaj 9839520b7b refactor(alarms): one date formatter, and a picker year range sized to its date
The list and editor now format a date through the same helper as the
widget, and the date pickers keep their default years unless a stored
date lies beyond them, instead of always spanning to 9999.
2026-10-22 08:20:19 +02:00
makiolaj beeca29f74 fix(tiles): redraw on clock and zone changes, and tighten the tile surfaces
The tiles' refresh now rides WidgetSync's collectors instead of a second
copy of them, and SystemEventReceiver asks both tiles to redraw on a time
or zone change, where the times they show move without any write.

The quick-timer tile remembers its timer before starting it, so a failed
start can't leave a timer it cannot see, and shows the timer's stored end
rather than one re-derived from two clocks on every read.

The tile preferences screen closes on anything but QS_TILE_PREFERENCES,
runs in its own task, and asks for a redraw after saving a duration. The
tiles and the next-alarm widget share one 12/24-hour time formatter and
the widgets' launch intent.
2026-10-22 08:19:49 +02:00
makiolaj 7adda73a79 fix(alarms): keep the editor's Delete off a ringing alarm too
With its confirmation gone, one tap on the editor's Delete would have
silenced a ringing alarm round its dismiss challenge, which the list's
swipe is already kept from doing; the ring screen is the way out. The
list's delete now runs on the application scope like the editor's, a
thrown-away card ignores a new drag, and the timers snackbar clears the
live pill on an empty tab.
2026-10-22 08:15:52 +02:00
makiolaj 5ad9bd02d6 fix(alarms): spend a pending skip only after its fire grace
The upcoming check that spends a skip dismissed in advance now wakes once the skipped occurrence's fire grace is over rather than at the occurrence itself, so it can never get ahead of a fire still in flight for it.

Fixes #16
2026-10-22 08:13:31 +02:00
makiolaj c5de9becdf fix(alarms): spend a date on the ring that used it, and harden date input
A dated one-shot whose time was moved later while it was snoozed rang a
second time that day, because closing the cycle only asked whether the
date was still ahead. It now keeps the alarm on only for a date after the
day that rang, and a one-shot switched off as its cycle closes lets its
date go.

The date pickers accept every year a stored date can have, so a far date
from a backup can no longer crash the editor. The widget formats its date
with the list's own formatter, and a backup's leftover date on the wrong
kind of alarm is ignored instead of rejecting the file.
2026-10-22 08:11:35 +02:00
makiolaj da9854abf8 feat(tiles): Quick Settings tiles for the next alarm and a quick timer
"Next alarm" reads the next-alarm widget's model and shows the weekday
and time of the next alarm, or "No alarms"; a tap opens the Alarms tab,
behind an unlock on a locked device, through the PendingIntent overload
of startActivityAndCollapse on API 34+.

"Quick timer" is the quick-timer widget's twin: at rest it shows its
duration and a tap starts a transient timer through the timer engine;
while that timer is active the tile is on, shows when it ends (or that
it is paused or done), and a tap resets it. It works on a locked device,
as the timer notification does. Its duration is a new backed-up setting,
chosen by long-pressing the tile, with the widget's bounds and picker.

Both are active tiles, redrawn by TileSync when the state they mirror
changes rather than bound on every panel open. No new permission.

Fixes #26
2026-10-22 08:09:14 +02:00
makiolaj fdb725e64c fix(alarms): spend a skip at its occurrence again
Keeping a skip armed through its fire grace let the editor's switch or another dismissal inside those two minutes ring the skipped alarm late, or be absorbed. The resolver goes back to spending the skip at its occurrence and recording it as handled, and the upcoming check wakes at an armed skip's occurrence so the list stops saying "skipping next" on time.

Fixes #16
2026-10-22 08:07:52 +02:00
makiolaj 1e29226364 feat(screensaver): add a screen saver with a dim night mode
A DreamService shows a full-screen clock, the date and the next alarm,
digital or as the world clock's analog face, on black. Night mode dims
every colour towards black and turns the display down; the clock moves
to a new spot each minute so nothing burns in. It wakes only on the
minute and on a time, zone or 12/24-hour change.

The tabs' settings gear becomes an overflow menu with "Screen saver" and
"Settings". An app cannot start the system's dream without a system
permission, so that item opens an activity with the same content, which
keeps the screen on only while charging. Style and night mode live in
Settings > Clock and in the dream's own settings screen, and are carried
in backups; older backups restore the defaults.

Fixes #25
2026-10-22 08:07:44 +02:00
makiolaj 5f3b97b32a fix(alarms): refuse a list delete of a ringing alarm in the engine
The list's delete re-checks for a ring under the engine lock, so a
swipe released as the alarm fires cannot silence it round its dismiss
challenge. An Undo whose snackbar is already up no longer goes stale
on a rotation after a long accessibility timeout, and a deleted card is
thrown clear of the screen edge.
2026-10-22 08:05:25 +02:00
makiolaj a88e696d52 fix(alarms): spend a skip only after its fire grace
The resolver now keeps a skip armed until its occurrence is out of the fire grace too, so a reschedule in those two minutes cannot spend it early and let a fire already in flight ring the dismissed alarm; onFire refuses that fire by the armed watermark. The notifier also cancels notices it posted itself, which the platform may not list as active yet.

Fixes #16
2026-10-22 08:00:38 +02:00
makiolaj 697678b226 fix(alarms): drop an Undo offered while no list could show it
A delete can finish while no list is on screen, such as the app sent
home the instant the editor's Delete was tapped. Its Undo is now
dropped unshown once it is 30 seconds old, instead of surfacing hours
later. Alarms and timers share one UndoSlot for this. A card whose
swipe is re-enabled mid-settle still slides home.
2026-10-22 07:54:43 +02:00
makiolaj 247dbcb751 fix(alarms): name a pause's last day, and switch on an alarm given a date
"Paused, back on" named the day after the pause, which for a weekday
alarm can be a Saturday it never rings on. A pause under way now reads
"Paused through Wed, Oct 28".

Picking a date on a switched-off alarm switches it on, the pause picker
cannot confirm a range that is already over, a spent pause stored with
its ends reversed is cleared too, and the widget formats its date in the
app's own locale.
2026-10-22 07:53:11 +02:00
makiolaj b4af164826 fix(alarms): spend a dismissed skip on time, and ring a re-armed handled alarm
The upcoming check now also wakes once a noticed occurrence's fire grace has run out, so a notice cannot outlive a late fire, and once an armed skip's has, so the list stops saying "skipping next" about an occurrence already gone. onFire is back to checking only the armed skip: the resolver deliberately re-arms a handled occurrence that a clock moved backwards put back in the future, and that one must ring.

Fixes #16
2026-10-22 07:51:57 +02:00
makiolaj b79b38bd2b fix(worldclock): fix Etc/GMT names at the fallback, not per caller
The previous rule named every non-place id by its current offset, which
renamed legacy ids such as Japan, called Etc/UTC "GMT" and made a CET
row's name change at DST. Only the Etc/GMT+-n ids have a misleading
last segment, so ZoneCatalog.cityFallbackFor now flips just those to
their real offset. That also reaches the zone picker's device row,
which takes its name from the same fallback.
2026-10-22 07:44:54 +02:00
makiolaj 62da327771 fix(alarms): bring a card home when its swipe is disabled mid-gesture
A card whose alarm starts ringing during a drag or the throw-away
animation slides back instead of staying displaced, and the delete is
re-checked after the animation, so it cannot land on a ringing alarm.
2026-10-22 07:44:28 +02:00
makiolaj 1dcd743e5b fix(alarms): keep a skip on the next ring after any edit, and say when a pause ends
Moving the time or changing the days left a pending skip pointing at the
old sequence, so the skipped day could ring at its new time. Every editor
change that can move a fire time now re-arms the skip.

A pause under way now reads "Paused, back on Thu, Oct 29" instead of
naming its last day, which read as the day it ends before. Reopening a
begun pause keeps its start, the date pickers take their bounds from the
clock when they open, and the dismiss chooser shows the scheduled date and
pause like the list does.
2026-10-22 07:43:30 +02:00
makiolaj f570bdefb7 fix(alarms): keep the upcoming notice off a locked screen below Android 12
The Dismiss action's unlock requirement is honoured only from Android 12, so on 10 and 11 the notice is now secret on the secure lock screen rather than dismissible by whoever holds the phone.

Fixes #16
2026-10-22 07:41:05 +02:00
makiolaj 40b4c9e2e4 fix(worldclock): name offset-set zones by their real offset everywhere
The Etc/GMT-n naming fix only covered the local row. A home that
follows a device set by UTC offset still put "GMT-9" under a UTC+9
face. The rule now lives in the one place every row and the face take
their city from.
2026-10-22 07:38:49 +02:00
makiolaj 1253c0fc55 fix(alarms): no fire for an occurrence already accounted for
A natural fire whose instant equals the skip or the handled-occurrence watermark is ignored, so a dismissal racing a reschedule cannot ring. Dismissing a repeating alarm persists its resolution before arming the skip, so an unspent old skip cannot absorb it. The deferred notice sync runs even if starting the ring throws, the swipe record is one synchronised read-modify-write, and it stays out of backups.

Fixes #16
2026-10-21 20:26:50 +02:00
makiolaj 471cbdabd8 fix(alarms): re-arm a skip when the pause changes, and clear spent dates in place
A skip armed under one pause kept its watermark when the pause changed,
so removing the pause left the alarm silent until the old pause's end.
The engine now re-arms a pending skip after a pause edit.

Spent dates and pauses are cleared with targeted, conditional updates
instead of a whole-row rewrite. SET_ALARM ignores a date or pause that
has already gone by when looking for an alarm to re-use, the editor
re-reads the clock when a date picker opens, and the next-alarm widget
shows the year for a date in another year.
2026-10-21 20:25:28 +02:00
makiolaj f46b7e1e37 fix(worldclock): keep the local row's day label and legacy city names
The local row prepends "Local time" to the offset and day, so on a
phone a one-line summary cut off the day ("Tomorrow"), the part that
matters most on a long trip. It now wraps to two lines. Only Etc/
offsets and bare abbreviations are named by their offset; a legacy
place id such as US/Pacific keeps its place name. The picker's home
selection now derives from the same settings subscription the tab
uses.
2026-10-21 20:24:24 +02:00
makiolaj af12b7bb34 fix(alarms): keep a ringing card out of the swipe, and Undo off the ViewModel scope
A ringing alarm's card no longer swipes or offers Delete: it leads to
the ring screen, and a swipe must not get round the dismiss challenge.
A drag cancelled before release (a ring screen opening on top) never
deletes. Undo restores on the application scope, since a claimed Undo
has no second chance. Closing a cycle and deleting share one teardown.
2026-10-21 20:22:25 +02:00
makiolaj 1333001aa9 fix(worldclock): name an offset-set local row by its real offset
A device set by UTC offset rather than by city reports an id like
Etc/GMT-9, which is UTC+9; the local row took its last segment as the
city and read "GMT-9". It is now named by its actual offset. LocalClock
takes the already-resolved home zone rather than resolving it a second
time, the tab reads both of its settings from one subscription, and the
local row falls back to the unknown-zone sentence for TalkBack instead
of an empty time.
2026-10-21 20:18:10 +02:00
makiolaj dedcbd5dfd fix(alarms): a consumed skip stays handled, and only a swipe hides a notice
A skip spent on its occurrence now becomes the handled-occurrence watermark, so a second resolution inside the fire grace cannot ring the skipped alarm after all. The upcoming notice records a swipe through its delete intent rather than inferring one, so a reboot or a blocked channel no longer hides it for good; its Dismiss asks for an unlock, and a Dismiss for a switched-off alarm clears its notice.

Fixes #16
2026-10-21 20:18:05 +02:00
makiolaj b1a45db54a feat(alarms): schedule an alarm on a date, pause a repeating one
A one-shot alarm can now ring on a chosen date, and a repeating alarm
can be paused for a range of days. Both are local dates: the date is a
floor and the pause a hole in the occurrence sequence, so DST and zone
changes resolve as for any other alarm. A missed date behaves like a
missed one-shot, and each clears itself once it can no longer change a
candidate. Picking a repeat day clears the date, as in Google Clock;
"skip next" stays the quick version of a pause.

The editor uses the M3 date and date-range pickers, which offer no past
day. The list, the next-alarm widget and the self-check read the
resolver, so they follow. The schema goes to v4 with a migration, and
the backup format to v2, which still imports v1 files. SET_ALARM no
longer re-uses an alarm that carries a date or a pause.

Fixes #21
2026-10-21 20:16:34 +02:00
makiolaj ef26d433e1 fix(alarms): let an unanswered Undo lapse, and make delete and restore atomic
A snackbar the screen leaves unanswered (another tab, Settings, the app
closing) now counts as expired, except across a rotation. Held back, it
could restore a snapshot from hours ago, ring a long-gone timer, or
land on top of a backup restored in the meantime. One shared
UndoSnackbarEffect does this for alarms and timers.

The engines' delete and restore run non-cancellable, so a caller
cancelled half-way cannot leave a restored alarm unscheduled. The
editor writes the label still in flight before it deletes, so Undo
brings back what was typed.

The swipe counts the touch-slop overshoot, keeps a card flung back
towards the start, slides a card home if its row is still there a
second after the delete, and no longer recomposes on every drag frame.
2026-10-21 20:13:09 +02:00
makiolaj 5b35ef65d4 fix(alarms): keep the upcoming Dismiss away from the fire
A fire broadcast already in flight when a skip was armed on its occurrence is now ignored, and a Dismiss that lands once the occurrence is due is stale. The notices are synced after the ring starts on the fire path, and a swiped-away notice stays gone across a process restart. A switched-off alarm answers NotFound.

Fixes #16
2026-10-21 20:09:47 +02:00
makiolaj f5388042c6 feat(worldclock): show local time automatically while travelling
Issue #24 asks for Google Clock's "automatic home clock": a temporary
Home row while the device is somewhere the time differs from home. In
Google Clock that works because the hero clock is local time. Clockula
is the other way round: the hero face is home (the stored home zone, or
the device's zone when none is stored), every row is read against home,
and with a stored home the face already shows home time on a trip. A
Home row would only repeat the face; what goes missing while travelling
is the local time.

So the feature is turned round. With a home zone stored and the device's
zone on a different UTC offset right now, the tab shows the device's
zone as a temporary "Local time" row between the face and the saved
cities. It is never stored, sits outside the reorderable column, has no
move or remove actions and does not count towards the 24-city cap. The
comparison is the current UTC offset, as Google's is, so Paris with
home in Berlin shows nothing. The row is hidden when a saved city is
already on the device's offset. It is rebuilt on every tick from the
same device-zone read the face uses, so it appears and disappears
within a tick of a time-zone change while the tab is open.

Home that follows the device can never differ from it. Settings > Clock
gets an "Automatic local clock" switch (on by default) under the home
zone row; while home follows the device the row is dimmed and its hint
says it needs a home time zone. The switch is stored as
automatic_local_clock and included in backups. Files written before this
change restore it as on.

Fixes #24
2026-10-21 20:06:13 +02:00
makiolaj 2feed2ee35 feat(alarms): swipe to delete, with Undo
An alarm card swipes away towards the end edge, or deletes from its
"Delete" accessibility action, and the list's snackbar offers Undo.
The editor's Delete drops its confirmation for the same Undo, routed
through a singleton so the list can answer it after the editor pops.
Timer cards get the same Undo for their corner delete.

AlarmEngine.delete stops a ringing alarm and clears a snoozed one's
notification under the engine lock; restore puts the row back under
its own id with its ring state and reschedules. A ringing alarm comes
back dismissed, never ringing; a finished timer comes back reset.

The swipe is hand-rolled so the end-to-start drag stays with the tab
pager.

Fixes #20
2026-10-21 20:02:00 +02:00
makiolaj e43339c386 feat(alarms): notify before an alarm rings, with Dismiss
Two hours before a natural occurrence (AOSP DeskClock's lead), a silent
"Upcoming alarm" notification offers Dismiss, and the alarm's row shows
the same Dismiss as a chip. Dismissing is dismissUpcoming's own branches:
a repeating alarm skips that occurrence, a one-shot is switched off.

The notices are synced at the end of every reschedule, after the real
registration, and a third AlarmManager slot on its own request code
wakes when the next window opens. The sync is wrapped so a notification
failure can never stop a ring. Dismiss names the occurrence and is
refused once the alarm is ringing or has moved, so it can't get round
the dismiss challenge. No challenge is asked before the ring, as in
Google Clock.

Fixes #16
2026-10-21 20:01:21 +02:00
makiolaj d70d0c78da test(worldclock): make the seconds tests mean what they say
The no-re-emission check now waits for the write to be readable before
asserting, and the seconds switch has its own settings test instead of
riding in the clock-defaults one. WorldClockSource.state takes its
refresh signal without a default, and ARCHITECTURE.md's slice count
and alignedTicks guarantee match the code.
2026-10-21 19:57:19 +02:00
makiolaj 78da94d775 fix(worldclock): leave a fitting margin and drop the zone-cache reset
The fitted seconds size keeps five percent spare, since hinted glyph
advances do not scale exactly. The receiver no longer resets the
process's default zone: that was speculative, and it changed global
state from a UI flow. The settings state takes the seconds flag without
a default, and a garbled KDoc sentence is fixed.
2026-10-21 19:48:30 +02:00
makiolaj 6239ca0477 refactor(settings): read a silence-after of an hour or more as Never
A stored value of 60 minutes or more now reads as Never, which rings that
long anyway. The picker no longer shows both "60 minutes" and "Never" for
the same window. Also drops a redundant null check in TimerRingPolicy.

Refs #17
2026-10-21 19:45:44 +02:00
makiolaj 17a44369b6 refactor(timers): settle timer vibration by copying the alarm default
Telling a fresh install from an updated one (package install times, the boot id, a string-encoded tri-state) was more machinery than the default is worth. Timer vibration is a plain boolean again, defaulting to on like the alarm default, and a startup copy writes the alarm default into it once, so no install's timers change on update. The copy runs in its own coroutine, off the engines' boot pass.

Fresh installs start with timer vibration on, matching alarms; the switch is one tap away in Settings > Timers.

Refs #22
2026-10-21 19:41:46 +02:00
makiolaj 1689794da6 fix(alarms): re-anchor a ring behind a moved clock, own the timer write
Fifth round of review fixes for the silence-after setting.

A ring whose start the clock has moved back behind is now re-anchored to
now, once, and stored. Before, the negative age lingered, and every later
re-measure gave the ring a fresh window. With that fixed, onAutoSilence
can safely re-arm again when the ring is still inside the window in
force, so a backstop queued behind a raised setting no longer ends the
ring early.

The timers' setting is now written through
TimerEngine.setTimerSilenceAfter, which records a session that has
already outrun the old window as lapsed, so raising the setting cannot
restart a timer that went quiet while no service pass was running. The
lapse rule and the lapsed anchor live in TimerRingPolicy, shared by the
policy and the engine.

Refs #17
2026-10-21 19:38:37 +02:00
makiolaj 9516dc1198 refactor(alarms): derive the alarm-set notice from the engine alone
The announcer now records only which alarm a user action set, and when.
What the snackbar says is derived from upcoming() while a screen shows
it, re-subscribed per announcement so the first read is from after the
write. That removes the second resolve path through the engine lock
(upcomingFor is gone), the publish lock, and the flash of a pre-write
countdown. The engine is observed only while an announcement stands.

Freshness is judged when a screen picks the announcement up, so one
already showing is not cut off half-way. Consuming matches a sequence
number. Latest action wins across alarms again: the sentence names no
alarm, so a settling announcement must never land on top of a newer
action on another one. Switching an alarm off announces nothing, and a
dismissal retracts before it skips.

Refs #19
2026-10-21 19:35:29 +02:00
makiolaj 71875e95e4 fix(worldclock): fit the seconds readouts without touching the default
The row's time is fitted only with seconds on, against both day periods
since an AM marker can be the wider one, and the style is built once by
the screen rather than per row. With seconds off the row is exactly as
before. The home readout stays on one line with seconds, shrinking to
fit instead of wrapping. SCREEN_ON is dropped: the tab cannot be seen
again without a resume, which already refreshes. §5 no longer claims a
clock change heals within one tick for free.
2026-10-21 19:32:44 +02:00
makiolaj 0b7af0d57b fix(alarms): let the backstop always close the ring it was armed for
Fourth round of review fixes for the silence-after setting.

onAutoSilence no longer re-arms a ring it measures as inside the window.
After a clock moved back, that check could push the backstop out again
and stretch the ring well past its window. Every writer of the window and
every clock change already re-arms the backstop under the engine lock, so
a backstop that fires has had its window, and it closes the ring as in v1.

A one-shot alarm the resolver switches off as missed also clears a
pending skip flag, as the resolver's other one-shot disable does.
SNOOZE_GRACE is its own ten-minute constant again, not tied to the
setting's default.

Refs #17
2026-10-21 19:28:09 +02:00
makiolaj 820317f9d9 fix(timers): tell a fresh install by its boot id and repair a garbled timer vibration
An empty store was a racy sign of a fresh install: the UI can write the store (the notification-permission flag) before the startup settle runs, and a fresh install would then vibrate its timers. The boot id is written only by the engines' pass, which runs after the settle in the same coroutine, and an Auto Backup restore brings the old device's.

Once settled, the startup pass costs one read and no package-manager call. A garbled stored value is replaced rather than left following the alarm default.

Refs #22
2026-10-21 19:24:52 +02:00
makiolaj 7a86e5488e fix(worldclock): refresh on resume and size the row's time once
With the tick a minute long, a language change left the city names in
the old language until the next minute; the screen now sends a refresh
on every resume. The row's time is sized once per format against a
wide sample instead of searched for every second, and ellipsizes as a
last resort rather than clipping. BackupSettings takes the setting
without a default, so a snapshot cannot drop it silently.
2026-10-21 19:22:47 +02:00
makiolaj 4cb4b6b6eb fix(alarms): keep clock changes and setting writes inside one window
Third round of review fixes for the silence-after setting.

A backstop that fires while the clock is still behind the ring's start
now closes the ring instead of re-arming in a loop until the wall clock
catches up. A clock change re-measures a ring in progress before it
reschedules: a clock moved back re-arms the backstop at most one window
from the moved clock, and one moved forward past the window closes the
ring.

The engine now owns the alarm setting's write. setAlarmSilenceAfter
stores it and re-measures the ring under one lock and without
cancellation, so a write can no longer land without the re-measure.

A snooze the resolver drops takes its "Snoozed until" notification with
it. SNOOZE_GRACE is derived from the default window, and the silence-after
picker moves to ui/settings, since it is not a per-alarm override.

Refs #17
2026-10-21 19:20:25 +02:00
makiolaj d2bf4349c5 fix(alarms): never announce an alarm that has since stopped ringing
A notice is now re-checked against the engine before a screen shows it,
so an alarm switched off or deleted by any path - the ring screen, a
notification, an intent, a restore - is never announced as set. A
retract takes the publish lock, so a publish already past its read
cannot put the notice back.

Settling is per alarm: switching one alarm off in the list no longer
cancels the repeat days still settling on another. The identical-words
dedupe is gone, since it could swallow a real confirmation.

The editor's writes that move a fire time, with their reschedule and
announcement, run on the application scope. Pick, OK, back could
otherwise cancel the reschedule half-way and leave a stored time the
system had not been given. upcomingFor() reads alarm and state under
the engine lock, so a fire cannot land between the two reads.

Refs #19
2026-10-21 19:18:19 +02:00
makiolaj adc5a3bdd6 fix(worldclock): read the whole home readout and catch up after sleep
The home readout gave TalkBack its minute-precision time as a
description, which replaced "Home", the city and the zone name; it is
the node's text now. The clock-change receiver swaps TIME_TICK, which
doubled every minute's work, for SCREEN_ON, which covers a delay
stretched by deep sleep, and resets the cached default zone on
TIMEZONE_CHANGED so the rebuild cannot read the old one. Doc and KDoc
drift fixed.
2026-10-21 19:14:08 +02:00
makiolaj fb875ca71c fix(alarms): measure every ring against one window rule
Second round of review fixes for the silence-after setting.

A fire now clears any stale ringingSince before it re-resolves the slot.
Before, the resolver's new one-shot close could switch off, or delete, the
very alarm that was about to ring.

A one-shot alarm displaced by another alarm's fire is switched off, as on
every other close of an undismissed cycle, instead of being armed for the
same time tomorrow.

onAutoSilence, onSilenceAfterChanged and the reboot resume now share one
rule. A ring still inside the window has its backstop re-armed, measured
from when it started, or from now if the clock has moved back behind that
start, so a clock change cannot stretch a ring past one window.

Refs #17
2026-10-21 19:12:18 +02:00
makiolaj fa5ce55503 fix(alarms): settle and show the alarm-set notice where it can't be lost
The repeat-day settle ran on the editor's scope, so leaving mid-settle
needed a flush that a cleared scope could still race. The announcer now
owns the settle on the application scope. Screens show the notice only
while resumed, so a cross-fade never shows it twice and a backgrounded
app does not use it up unseen. The editor shows only notices about its
own alarm, since the sentence names none.

Deleting an alarm, or the AlarmClock chooser dismissing it, takes its
notice back. Freshness is measured on the monotonic clock, one instant
is used for both ends of the countdown, and an identical countdown
keeps the standing notice instead of flickering it.

Refs #19
2026-10-21 19:10:42 +02:00
makiolaj 65c86af313 fix(worldclock): keep seconds quiet for TalkBack and apart from the defaults
The time each row and the home readout give TalkBack stays at minute
precision, and the ticking text is kept out of the tree, so a focused
row is not re-read every second. The row's time is capped in width and
shrinks to fit, so a large font scale cannot crowd out the city name.

The setting is its own SettingsPrefs flow rather than a ClockDefaults
field, so flipping it re-emits nothing alarms and timers watch. The
clock-change receiver sits outside the flatMapLatest and restarts the
tick, registering once per subscription and re-measuring the wait
against the clock as it now reads.
2026-10-21 19:07:19 +02:00
makiolaj 84d31fe7f8 fix(timers): settle timer vibration outside the store's init and resume a ramp
The DataStore migration ran its write inside the store's initialisation, so a failed write (a full disk) would have failed every preference read, the alarm path's included. The timer-vibrate key is now nullable instead: unsettled reads as the alarm default, which is what timers did before, and ClockulaApp settles it once at the head of its startup pass, off only for a never-updated package with an empty store. A failed settle leaves the fallback in place.

A timer ring restarted mid-session (handed to a second expired timer, back after an alarm held it, a recreated service) now picks its ramp up from the session's elapsed time instead of either ramping again from a whisper or jumping straight to full volume. RingAudioPlayer takes the offset; alarms pass none.

The Vibrate row's switch is display-only, so a screen reader stops on the row once.

Refs #22
2026-10-21 19:06:50 +02:00
makiolaj 1c323d5b4c fix(alarms): apply silence-after mid-ring and close missed one-shots
Review fixes for the silence-after setting.

Changing the alarm's silence-after now re-measures a ring in progress at
once, through AlarmEngine.onSilenceAfterChanged: past the new window it
closes as missed, inside it the backstop moves. onAutoSilence also re-arms
instead of closing a ring that is still inside the current window, so a
backstop armed for a shorter window cannot end a ring early.

A cycle the resolver closes as missed (a ring past its window, or a
snooze that never fired) now disables a one-shot alarm, as the engine's
own close does. A short window made that path much easier to reach, and
the alarm was being armed for the same time the next day.

A late snooze is kept for the window or ten minutes, whichever is longer,
so a short window never makes a promised snooze easier to lose than in v1.

A timer that has gone quiet stays quiet when the setting is raised: the
lapsed anchor moves back past the hard stop. The timer's heads-up is now
keyed to the session anchor rather than the window's end, so editing the
setting mid-ring no longer counts as a new session and alerts again.

A negative stored value now reads as the default instead of Never. The
picker offers a stored value that matches no choice, and the wake-lock
timeout is one shared constant.

Refs #17
2026-10-21 19:04:08 +02:00
makiolaj 207475ef37 fix(alarms): keep the alarm-set notice until a screen has shown it
The usual way to set an alarm is pick, OK, back, and back disposed the
editor's snackbar a moment after it appeared. Day taps left within the
one-second settle were never announced at all, and a rotation threw a
showing snackbar away for good.

The announcer is now an app-wide singleton holding the notice as state
until a snackbar has run to the end, so the list picks up what the
editor could not finish showing. An editor closed mid-settle announces
on the way out. Announcing is serialised, so a quick on-then-off cannot
land the "on" read last, and a notice nobody showed within ten seconds
is dropped rather than shown stale. The engine's one-alarm read now
shares upcoming()'s preview helper.

Refs #19
2026-10-21 19:00:56 +02:00
makiolaj a4cec23e5c fix(worldclock): keep the seconds option from costing reorder and catch-up
Rows that tick every second re-seeded the reorderable column mid-drag,
so it is handed ids now. A zone or clock change no longer waits for the
next minute: TIME_SET, TIMEZONE_CHANGED and TIME_TICK are merged into
the tick through a new ClockChanges seam. With seconds on, the row's
time drops a size so the city name keeps its room. The truncation unit
and the tick come from one place, and stale 200 ms comments are gone.
2026-10-21 18:57:10 +02:00