omarchy plugin update pulls the new files and calls rescanPlugins, but that
does not reload a keepLoaded panel plugin -- the shell keeps the QML component
it compiled at startup. The files change and the bar goes on running the old
build, which reads as the update having silently done nothing.
Verified by writing a deliberately visible change straight into an installed
plugin directory: after rescanPlugins the bar still rendered the old colour,
and only omarchy restart shell picked it up.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MhxtL7WzL5XmuTd8GHbcTY
Finding a zone meant already knowing its name, which is the wrong way round
when the thing you know is the country -- Ghana's zone is named after Accra,
and nothing in the picker said so.
Zones now carry the countries they cover, drawn from tzdata's own tables:
zone1970.tab and zone.tab for the zone-to-country-code rows, iso3166.tab for
the names. The picker already matched on description as well as label, so
putting the countries there makes them searchable and shows them in the list.
Those tables only name about 420 of the ~600 zones; the rest are
backward-compatibility aliases like America/Buenos_Aires and US/Pacific, which
tzdata ships as hard links to their canonical zone. Matching an alias to its
canonical name by inode lifts coverage to 553, and the 45 left are the Etc/GMT
pseudo-zones, which have no country by definition.
The inherited countries are the union of every zone sharing the inode, not the
first one found: those siblings are one clock under several countries' names,
and each table names a different subset, so first-match made Africa/Asmera
findable as Ethiopia but not as Eritrea.
Joining is done in ClockModel against a fixture, so it does not depend on the
host's tzdata version; the shell side only reads the files.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MhxtL7WzL5XmuTd8GHbcTY
Typing a label and clicking Save straight away saved the old value. Fields
commit on editingFinished and spin boxes parse their typed text on focus loss,
so a value that had been typed but not blurred was never in the draft that
Save read -- and the typed text went with the delegate when the draft array
was replaced.
Committing on every keystroke is not the fix: the draft is a var array, so
replacing it re-creates the row delegates and would throw away the field being
typed into. Instead every action that reads or replaces the draft -- Save, add,
remove, the timezone picker, the per-clock seconds toggle -- now moves focus
first, letting each control commit through its own normal path. Focus lands on
the panel's key scope, so Escape still closes the editor.
Verified against the reported case: at the moment Save ran the draft still held
the old label, and the file received the typed one.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MhxtL7WzL5XmuTd8GHbcTY
Seconds move from one bar-wide switch to a per-clock setting. A clock's own
`showSeconds` wins, the top-level one is the default for clocks with no
opinion, and the centre clock gets `local.showSeconds` since it has no entry in
`clocks[]`. An explicit false on a clock beats a true default, so the check is
for the property's presence rather than its truthiness.
The tick rate follows what is actually displayed: `anyShowsSeconds()` decides
between a per-second and a per-minute SystemClock, so a bar showing no seconds
does not wake the shell sixty times as often to redraw an identical minute.
`barSize` is now editable from the popup, and everything drawn on the bar is
sized against the theme's own bar height rather than pinned to the theme's
12px type. A taller bar carries proportionally larger text, gaps and rule; at
the theme's own height the scale is exactly 1, so nothing moves. Heights are
clamped to 12-120 rather than rejected, so a value the editor can produce can
never be one the config layer silently discards.
Fixes a latent bug this surfaced: updateRow() rebuilt a draft row from a fixed
pair of fields, so editing a clock's label would have thrown away its own
seconds setting. It now copies every key the row had.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MhxtL7WzL5XmuTd8GHbcTY
A `panel` shell plugin drawing its own bottom layer-shell surface, one per
screen, with an exclusive zone so Hyprland tiles above it. Clocks are laid out
as a single timeline around local time: side and position derive from each
zone's live UTC offset, so config order is irrelevant and a DST shift re-sorts
the bar on its own.
Qt's QML engine ships no `Intl`, so an IANA zone cannot be rendered from inside
QML at all. Offsets come from the system's own tzdata via `date +%z` and the
arithmetic is done here, which stays correct across DST and tzdata updates
without encoding any of the rules. The probe runs on a slow timer and on hour
boundaries; the minute tick advances the clocks in between.
Config lives in ~/.config/omarchy/clockbar.json and is edited either by hand or
through a click-to-edit popup with a timezone picker. Both routes write the same
file and the bar reloads from it, so they cannot disagree.
Failures are shown rather than papered over: an unresolvable zone renders
`LABEL --:--` in the urgent colour instead of silently falling back to UTC, and
an unparseable config reddens the bar's hairline and flags the local clock while
the last good config keeps ticking.
Ordering, formatting, day offsets and working hours live in ClockModel.js, which
touches no QML types and is covered by tests that run under node.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MhxtL7WzL5XmuTd8GHbcTY
