A machine without a GPU, like most VMs, renders through llvmpipe on the
CPU, where every animated frame and every translucent window costs. In a
VM, opening and closing a terminal took ~5s of CPU; a panel ~3.4s.
omarchy toggle animations (also under Toggle > Animations) places a Hyprland
flag that turns off animations, blur and shadows and makes windows opaque.
The shell follows Hyprland's animations:enabled, rereading it on every
config reload: its one-shot animations run for Style.duration(ms), which
is then 0, and its spinners, pulses and title marquee hold still. A new
install in a VM starts with the flag in place.
Measured in the ISO test VM (llvmpipe), CPU per interaction:
terminal open+close 5000ms -> 481ms, workspace switch 1670ms -> 409ms,
audio panel 3446ms -> 796ms, volume OSD 2324ms -> 584ms, menu 2028ms ->
1241ms.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Answer omarchy-shell calls over the shell's own socket
Every omarchy-shell call started a qs ipc client, ~45ms of startup for one
IPC call: a theme switch makes two, and every script-driven OSD, toggle
refresh and lock query paid it too.
The shell now serves a socket in XDG_RUNTIME_DIR, named from its config
path and Wayland display as qs ipc selects its instance, and omarchy-shell
tries it first through socat, which starts in ~5ms. First-party handlers
register as ShellIpc, an IpcHandler that qs ipc still reaches, and the
socket calls only the functions a handler declares with their exact
argument count, allowed by name so QObject methods such as destroy() stay
out of reach.
When the shell ran nothing it answers SKIP, and omarchy-shell asks qs ipc
for its exact answer, so errors, third-party plugins and an unreachable
socket behave as before. A call that may have run is never retried: a
timeout or a connection closed without an answer reports the shell as not
responding.
omarchy-shell shell ping takes ~13-18ms instead of ~61ms, and omarchy-osd
reaches the screen in ~36ms instead of ~77ms. Output and exit status match
the qs ipc path across 26 calls, errors and quiet mode included.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Only accept whole socket replies and retry only unmade connections
A reply cut off after its OK prefix passed for the whole answer, and an
empty reply with socat failing was retried through qs ipc although the
request might already have been delivered.
An answer now counts only once its record separator arrived. socat's own
errors join the reply, so only its connect error, a socket nothing
listens on, falls back to qs ipc beside an explicit SKIP; anything else
is reported as not responding rather than retried.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
The root rule matched only a file-level root Text, of which this tree has
exactly one. QML inline components are roots for the same reason — the
`text` of `component InfoValue: Text {` comes from every caller, so the
file it lives in never binds it — but they sit inside another element, so
the depth-1 test never saw them. Six went uncovered while the test
reported green, among them the network panel's InfoValue, which callers
bind to the IP address and gateway.
Six more ways to write a Text were read as clean rather than as unreadable:
an opening brace that is not last on its line, a brace on the line after
`Text`, a one-line block containing nested braces, a wrapped binding split
by a comment or a blank line before its `+` (which exempted a dynamic
binding as a literal), and a root Text indented from column zero. Require
the forms a line scanner can read instead of parsing QML; the tree already
writes every Text that way.
Last, a run that read no files reported success. A checkout with no shell/
QML now fails instead, since an all-clear from a scan that opened nothing
is the one answer this test must never give.
Each case is covered by a fixture that fails without its fix.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: OpenAI Codex (gpt-5, xhigh) <noreply@openai.com>
* Add Tailscale header on/off toggle
* Address Tailscale toggle review feedback
* Hide Tailscale tooltip while busy
* Keep Tailscale labels clear of header toggle
* Extract the switch from Toggle into a reusable ToggleSwitch
Toggle rendered its own track and knob inline, so anything else wanting a
switch had to copy the geometry. Pull it into Ui/ToggleSwitch.qml and let
Toggle compose it, and give PanelHero a trailingControl slot so a hero can
pin a control to its trailing edge without the caller doing the layout.
The switch draws its cursor as a ring outside the track: themes give normal
chrome a stronger border than hover-cursor, which is right for controls that
are borderless at rest but would make a bordered track go fainter under the
cursor.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Keep the panel cursor on the virtual header section
clampCursor resets focusSection whenever it is not in visibleSections, but
"header" is virtual and never appears there. Any refresh of the underlying
model therefore threw the cursor off the hero toggle: muting republishes the
PipeWire snapshot, and toggling the Bluetooth adapter empties and refills the
device lists. moveCursor already special-cases "header"; clampCursor now does
too.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Toggle panel heroes with a switch instead of the status icon
The hero icon doubled as the on/off control, which was invisible as an
affordance and made the icon carry two jobs at once. Give Tailscale, Dropbox,
Bluetooth, Audio, and Network a ToggleSwitch on the trailing edge of the hero
and leave the icon to report status. The switch is the header's only cursor
target, so the keyboard reaches it the same way the mouse does.
Dropping the icon's focus ring also drops heroRingPad, which lets each hero
line up with the rows beneath it. Network's link detail moves inline after the
name -- "Ethernet (2.5gbit)" -- since the pill crowded the switch.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Make the network hero switch honestly a Wi-Fi switch
The switch reads and writes Networking.wifiEnabled, but its tooltip claimed to
turn "network" on and off whenever Ethernet was the active connection. A click
asserted nothing; a switch asserts state, so on a wired machine with the radio
off it sat there reading "off" beside a perfectly live Ethernet connection.
Say Wi-Fi, and only offer the switch when there is a radio to switch.
headerActionCount follows the same condition so the keyboard cannot reach a
control that is not there.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Stop the header cursor claiming an absent switch
The switch hides when the thing it toggles is unavailable -- no Tailscale CLI,
no Dropbox CLI, no Wi-Fi radio -- but "header" stayed reachable, so the cursor
could sit on a target that never rendered. The old clickable icon was always on
screen, so there was always something to highlight.
"header" stays navigable and Enter still no-ops safely; the cursor just stops
claiming a spot that is not there.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Wrap the gallery's switch caption
The caption sat unbounded inside the switch row. The gallery has a 560px
minimum width, horizontal scrolling off, and clipping on, so at that size the
end of the line was simply unreachable. Move it below the row and wrap it, the
way every other description in the gallery already does.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Each panel's hero on/off toggle (Tailscale, Dropbox Wi-Fi, Bluetooth,
network, audio mute) now has a reachable "header" cursor section, a focus
ring around the hero icon, Enter/Space activation, a letter shortcut, and
hover parity. The hero is inset by heroRingPad so the ring stays inside
the Flickable/ScrollView clip box instead of being cut off.
Bluetooth and audio get a virtual "header" section above their device
sections, so the adapter/mute can be toggled by keyboard even when no
device rows exist. Network repurposes its previously dead header section.
For the PanelHero-based panels (Tailscale, Dropbox), route panel state
through the wrapper's `header` id: inside a PanelHero iconComponent `root`
resolves to PanelHero, not the Panel, so `root.headerHasCursor` and
`root.setHeaderCursor()` silently referenced the wrong object.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Click the Dropbox hero icon to pause or resume the daemon (dropbox-cli
stop/start), with a tooltip on hover. When paused, both the hero and the
bar icon grey out and the phrase reads "Syncing paused". The greying is
optimistic so it reacts the instant you click, then reconciles with the
real daemon state once it settles (reverting if the command failed).
Also fix status.py's running detection, which was stuck reporting true:
this dropbox-cli inverts the `running` exit code and reports "isn't
running" (not "not running") when stopped, so the paused state never
showed. Detection now keys off the status text alone.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
