Phase 4, first slice. Theme.qml hardcoded the Prism pair, so the one
thing that carries every state meaning in the desktop -- focused, active,
on -- was the one thing nobody could change. 74 files read Theme.accent,
so making it a setting moves all of them at once.
Named accents rather than a colour picker, which is the design decision
worth defending. One hex cannot serve both schemes: a colour legible on
the Moon background is usually illegible on the Day one, and a picker
that lets someone build an unreadable desktop is not a feature. So each
name carries a curated pair per scheme, and every one of the sixteen
resulting colours measures at least 3:1 against the ground it sits on --
checked, not assumed. It is also GNOME's model, which is the parity
being chased.
The focused window border comes with it, and only because the ownership
rule made that safe. ColorScheme owns the inactive border as a
scheme-relative contrast role; the focused Prism border is the accent
role owned by the theme. Writing it from the accent would have been
reckless before that boundary existed, since a scheme change would have
erased the user's choice. Both borders are now pushed together, because
each accent carries separate light and dark pairs, so switching schemes
must restate the focused border too.
The gradient is written as a Lua table, not a string. The string form
carries only one stop, and passing two as a string is accepted and
silently keeps the previous value.
Swatches are drawn as the gradient they produce rather than as flat
dots, because the gradient is what is being chosen. Each carries its
name permanently rather than in a tooltip: telling swatches apart by
colour is precisely what someone with a colour vision deficiency cannot
do, which is also why the palette is named in the first place.
Verified end to end by switching to rose and watching the compositor
report eeff757f/eec099ff, then reverting.
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
Super+Tab already cycled windows, but nothing was drawn, so you chose
blind and could only confirm the choice by arriving. A visible switcher
is muscle memory for anyone arriving from macOS or GNOME, and it was the
last item of roadmap phase 03 that did not need coordination.
Ordered most-recently-used, not by creation, because that is what makes
the gesture useful: one Tab returns to the window you just came from.
Hyprland does not report an MRU order, so it is tracked from focus
changes and keyed by address, which is the only property stable for a
window's lifetime.
The gesture needs three binds rather than two. Tab steps the selection,
and the switch is committed on Super RELEASE -- the only way the
compositor can say the gesture is over. That bind is on the bare
modifier, so it fires on every Super release in the session; commit()
returns immediately when nothing is open, which is what makes it
affordable.
A list of names rather than thumbnails: at a glance you are looking for
"the other terminal", and a row of live previews is slower to read and
far more expensive to draw than this gesture deserves.
The interesting part is the bug. The overlay was built, mapped nothing,
and logged absolutely nothing -- because it declared `required property
var screen` while Variants supplies `modelData`. shell.qml has carried a
comment warning about exactly this since the Bar hit it, and I read that
comment earlier in the same session and still walked into it. A comment
that does not stop the person who read it is an argument for a test, so
per-screen-surface-contract now checks every per-screen delegate takes
its screen from modelData. Verified it catches the exact mistake.
Also fixes a regression from 8be3fc2: settings-pages-contract still
required vitalsIntervalMs on Home, where it no longer is. That contract
was pinning the split-across-two-pages arrangement the same commit
fixed, and I pushed without running it.
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
Four places in the entire shell could reach Settings. The bar, where a
person looks first, was not one of them -- and Pill has routed
right-click to a secondaryActivated signal all along, which nothing
connected, so the gesture did nothing on every widget in the bar.
Each widget now opens the page that owns its settings: the clock and the
calendar reminder open Date & Time, weather opens Home, the vitals
readout opens Appearance, the status glyphs open Network & Devices, the
media readout opens Sound, and the privacy indicator opens Privacy &
Security. Left-click behaviour is untouched.
Two routing bugs found while picking those destinations, both of the
same kind and both invisible from the code, since each page reads
perfectly well on its own:
weather routed to Appearance while every weather control lives on
Home, so searching "temperature unit" opened a page without it.
vitals routed to Appearance, but the refresh interval sat on Home
while the toggles it governs sat on Appearance -- one concept split
across two pages, which is exactly what the ownership rule forbids.
The interval now sits beside the toggles and Home's stub card is gone.
The jump contract guards the failure mode these share. openSettings()
falls back to Home for an unknown page, sensibly and completely
silently, so a typo or a later rename turns a right-click into "opens
the wrong page" with nothing logged. It also fails a Pill-based bar
widget that leaves right-click unconnected, since that is how the
gesture came to be inert everywhere in the first place.
A third instance of the routing bug is still open: followMouse and
pointerSensitivity sit in the input group, which routes to Keyboard,
while both render on Mouse. Fixing it is a two-line group change in
PreferenceSchema.qml, which codex currently owns, so the contract that
catches all three lands with that fix rather than red.
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
Measured the gap first: of the 38 real Hyprland options Panama's own Lua
sets, only 16 were editable in Settings. Everything else required a text
editor, which is the thing this app exists to stop. This closes most of
that: 66 mapped options now, from 47.
Window shape and shadows on Appearance: corner shape (rounding_power),
focused and fullscreen opacity, shadow falloff and hard-edged shadows.
Window edges, master layout and Hyprland's own notices on Desktop & Dock.
Three of these are corrections rather than additions.
Master layout options existed nowhere, while Settings has offered "Master
and stack" as a choice since this morning -- a layout you can select and
cannot configure is barely a choice. Its card is hidden unless that
layout is actually selected, since settings that do nothing under the
layout you are running are worse than not offering the layout at all.
The four Hyprland notices -- logo, splash, update news, donation nag --
are all turned off by looks.lua on the user's behalf. Defensible as a
default, but not a decision anyone could reverse. They are stored
positively ("show this") and written as Hyprland's `disable_*` through a
new `invert` flag, because a switch labelled "Disable splash text" that
must be ON to hide something is a small cruelty. The Lua does the same
inversion so both sides agree.
Everything new also reads from prefs in looks.lua. Without that these
would apply live and silently revert on the next compositor reload,
which is the failure this codebase keeps designing against.
Two shapes the write path had never seen. Border colours are gradients
and shadow offsets are vec2, and the verifier understood neither -- it
returned false for anything outside int/bool/float/str/css, so both
would have reported every write as rejected. Gradients also need real
care: the stubs declare them as `string|{colors,angle}`, and the string
form carries only ONE stop, so writing "rgba(a) rgba(b) 45deg" as a
string is accepted and keeps the previous value. Verified that directly.
They are also written in one notation and read back in another
(`{colors={"rgba(3b426199)"},angle=45}` becomes `993b4261 45deg`), so
comparison normalises both sides.
Border COLOUR is deliberately not exposed yet. col.inactive_border is
written by ColorScheme on every scheme change, so a user's choice would
be silently overwritten, and col.active_border is the Prism gradient,
which needs a colour control this app does not have. Shadow offset is
left out for the same reason -- the vec2 support is in place for
whenever the widget exists.
Verified each new option applies and reverts against the live
compositor, and that the schema, enum-map, nav, write and commit/reset
contracts all pass.
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
Brings in panama-doctor (a 25-check diagnostic with fixture-backed
contracts), a Health service, a System Health page replacing Startup &
Services, and a bar indicator that stays absent until something is
actually degraded. All seven of its contracts pass on the merge.
Three things needed resolving rather than accepting:
The branch predates the debranding, so its user-visible strings still
named the product -- "Panama desktop is healthy", "Restart Panama",
"Panama tools". Rewritten to say the same thing without the name, which
is what the rest of the app now does.
Its Fedora hand-off card was a single button calling openGnomePanel
("network") under a subtitle naming five subjects. Main had already
replaced that with a row per subject, each opening the panel that owns
it, so those rows are ported into HealthPage instead. Printers and
online accounts stay on Network & Devices with the rest of the network
hardware.
That broke its own assertion, which matched the literal
openGnomePanel("network") string. Rewritten rather than reverted: it now
checks the boundary card exists and that every panel named in HealthPage
is one openGnomePanel actually allows, since a name outside the
allow-list opens nothing at all. Verified it catches a plausible-looking
wrong name.
SettingsShell and SettingsSidebar conflicted because both sides added
pages; resolved as the union, keeping its System Health page and live
footer alongside main's Mouse & Touchpad, Privacy & Security, Region &
Language and Online Accounts.
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
GNOME's Wi-Fi panel has this and it is the most-used thing in it: the
alternative is reading a passphrase out loud. Network & Devices now shows
a scannable code for any saved network whose passphrase this user can
read.
The image contains the network password in machine-readable form, so
most of the care here is about that rather than about QR codes. It is
written under XDG_RUNTIME_DIR -- 0700, on tmpfs, gone at logout --
rather than /tmp, which is shared; the file is 0600; the passphrase is
piped to qrencode on stdin rather than passed as an argument, because
argv is world-readable through /proc for as long as the process runs;
and it is never printed or included in an error message. Generated on
demand, because producing a code for every saved network up front means
writing images of passwords nobody asked to see.
Enterprise networks are listed as not shareable rather than offered and
broken: there is no passphrase to encode, so the code could not work.
Two bugs the contract caught while being written. Semicolons in an SSID
were not escaped -- the sed replacement had one backslash where its four
neighbours have two, so sed dropped it, and an SSID containing a
semicolon would have produced a QR code describing a different network.
And nmcli's trailing newline landed inside the payload; it decoded here,
but a newline in the middle of a WIFI: URI is not something every phone
tolerates, and that failure would present as "the QR code just doesn't
work on my phone".
The contract stubs nmcli and qrencode, because the real ones would write
this machine's actual Wi-Fi password into a fixture directory. It
asserts the escaping, the absence of a newline, the file and directory
modes, that no temporary payload survives, and that the passphrase never
reaches argv.
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
The same Power Saver / Balanced / Performance choice GNOME's Power panel
offers, and the daemon behind it was already running here -- it simply
had no control anywhere in Panama. This machine has been sitting on
"performance" with nothing to say so.
Talks to the net.hadess.PowerProfiles interface rather than to a binary.
Fedora 44 implements it with tuned-ppd instead of power-profiles-daemon,
and powerprofilesctl is not installed at all, so anything shelling out
to that command would have found nothing while the service was right
there. Setting a profile needs no privileges: the daemon accepts a
property write from the active session user.
Not a stored preference. The daemon owns the profile, it survives Panama
restarts, and anything else on the system can change it, so a copy in
settings.json would drift -- the same reasoning as monitor brightness.
PerformanceDegraded is surfaced because it is what makes the setting a
lie: a thermally throttled machine reports "performance" while behaving
otherwise, and that is worth saying out loud.
The contract stubs busctl, because the real daemon is a system service
shared with everything else on the machine and a test that flipped the
daily driver to power-saver and then died would leave it there. It pins
the parsing in particular: busctl renders the Profiles property flat, so
profile names and driver names arrive in one stream, and a pattern loose
enough to match both reports the driver as an extra profile. On this
machine the driver is called "tuned", which reads exactly like a
plausible fourth profile -- verified the contract catches that, and
catches an unvalidated name reaching the system service.
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
GNOME Online Accounts is a daemon plus a D-Bus API, and the daemon
already runs in this session -- gvfs activates it, and all four accounts
on this machine work without gnome-shell involved anywhere. Only the
PANEL was GNOME's. The accounts themselves are ordinary D-Bus objects
that anything may read and modify.
So everything except the initial sign-in is now native: the account
list, per-service toggles for mail, calendar, contacts, files, photos,
music and chat, and removal. That is the whole Online Accounts panel
apart from one OAuth handshake.
Signing in is the exception, and only for OAuth providers. The daemon's
AddAccount takes credentials as an argument -- it stores them, it does
not obtain them -- and the code that runs Google's OAuth exchange lives
in libgoa-backend, which Fedora ships without a GIR binding, so it is
reachable from C only. Reimplementing it would mean our own Google
client credentials. That step is handed to GNOME's panel and the page
says so, because a hand-off the user does not expect reads as a bug.
Password-based providers (Nextcloud, IMAP, WebDAV) could be added
natively later; their credential keys are known now.
Accounts needing re-authentication are surfaced first, which turned up
something immediately: both Google accounts on this machine report
attention_needed, meaning their tokens have expired and they have
stopped syncing. GOA has known that all along and nothing outside its
own panel ever said so.
Every write re-reads the account list rather than assuming it landed.
GOA can refuse, and a toggle that springs back is the honest outcome.
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
The keyring is already unlocked at sign-in exactly as GNOME does it --
pam_gnome_keyring is in GDM's stack and the journal confirms it works
("gnome-keyring-daemon started properly and unlocked keyring"). So there
was no configuration bug to fix. What a bare Hyprland session lacks is
anywhere to see when that has stopped being true.
It stops being true rarely and expensively. gnome-keyring-daemon crashed
once on this machine -- an upstream abort in service_method_open_session,
with a core dump -- and D-Bus then activated a replacement. That
replacement never received the login password, so the keyring was locked
in the middle of a session that had unlocked it correctly at login.
Nothing announces this. What you see instead is a mail account that will
not authenticate, a git push that cannot find its key, or an integration
reporting "not configured", none of which mention keyrings. That is the
same root cause as the Home Assistant token failure earlier.
Privacy & Security now shows the state, offers an Unlock action that
raises the standard password dialog, and reports when the daemon holding
your secrets is a D-Bus replacement rather than PAM's -- because a
replacement that is currently unlocked was unlocked by hand and will not
survive a restart. The password never passes through Panama.
The contract stubs the secret service rather than touching the real one:
locking the login keyring breaks every saved password on the machine and
can only be undone by typing the password into a dialog, so it is not
something a test suite may do to a daily driver. Verified it catches a
helper that misreports locked as unlocked, and one that crashes instead
of reporting a missing service.
Worth recording: a locked keyring makes a NON-INTERACTIVE caller appear
to hang. It is not hung -- it is waiting on a dialog nobody is looking
at, which is exactly how the earlier secret-tool investigation lost an
hour.
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
Reported symptom: Electron applications and a Chromium-based browser,
all set to follow the system, went light when the desktop went light and
never came back. Nothing reported an error, and the portal was serving
the correct value the whole time.
Cause: ColorScheme set gtk-theme to "Adwaita-dark" for dark and
"Adwaita" for light. Neither is installed on Fedora 44 -- only adw-gtk3
and adw-gtk3-dark are. GTK responds to an unknown theme name by falling
back to its built-in default, which is LIGHT. So asking for light worked
by accident, asking for dark silently produced light, and anything that
takes its cue from the GTK theme rather than the portal stayed light no
matter what org.freedesktop.appearance said. Verified: the portal emits
correctly in both directions, so this was never the portal's fault.
Second cause, the mirror of the first: gtk-3.0/settings.ini and
gtk-4.0/settings.ini were pinned to adw-gtk3-dark and prefer-dark=1 and
never regenerated. Under GNOME that file is ignored because
gnome-settings-daemon publishes over XSETTINGS; under Hyprland nothing
does, so for GTK3 it is authoritative -- and it contradicted the scheme
in light mode. They are now generated from a template on every switch,
gitignored as machine state, and seeded by link-dotfiles, the same shape
kitty's current-theme.conf already uses. These directories are symlinked
into the repository, so writing the live file directly would dirty the
working tree on every theme switch.
The failure was silent by construction, so it gets a test rather than a
comment: gtk-theme-contract asserts every theme name Panama sets is
actually installed, and that the generated GTK config agrees with the
scheme in both directions. Verified it catches both original bugs.
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
Two things, both closing gaps in work that was already reported done.
GNOME's Multitasking panel, in Hyprland's terms: tiling layout, split
behaviour, floating-window snapping, workspace wrap-around and
back-and-forth, whether applications may take focus, and whether the
pointer changes the active display. Hyprland creates and destroys
workspaces as you use them, so there is no fixed count to expose, and
the page says so rather than leaving a conspicuous absence.
The Desktop page described the first two of these as read-only facts --
"Layout: Tiling", "Workspace movement: Dynamic" -- which was never true.
Both are ordinary Hyprland options that happened to have no controls,
and TextRow's own documentation says a setting the user could reasonably
change does not belong in it. Schema defaults are Panama's shipped
values from looks.lua rather than Hyprland's own, so restoring defaults
returns the desktop to how it ships. 43 mapped options now, from 35.
The launcher had no light theme. vicinae.json already selected a theme
per system appearance, but both entries pointed at Moon, so choosing
light mode left the most frequently opened window on the desktop dark --
a hole in the light/dark work, not a missing feature. Day is authored
from the same palette as the kitty Day theme so the two cannot drift,
and link-dotfiles now installs every authored theme rather than only the
dark one, which is why the gap survived being noticed.
Its placeholder colour is not Tokyo Night Day's own: that measures
2.54:1 against the background, below the 3:1 floor for secondary text.
This is 3.25:1, the same value used for neovim's light comments.
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
More GNOME Settings parity.
Region & Language is new. The locale is localectl's, and Panama stores no
copy of it -- there is exactly one system locale, so a preference here
would be a second source of truth that drifts the moment anything else
changes it. Codes are resolved against iso-codes into "Portuguese
(Brazil)" the way GNOME does, with the code kept visible because it is
what actually gets written and someone choosing between two Spanish
variants needs to see it. Changing it is privileged and only applies to
programs started afterwards, so the page says a sign-out is needed
rather than claiming the new language is in use.
The service is called SystemLocale, not Locale: QML has a built-in
Locale value type that silently shadows a singleton of that name, and
every binding then reads properties off the wrong thing. The page
rendered empty with nothing but "cannot read property of undefined" to
explain it.
About now answers what GNOME's About answers -- model, processor,
memory, disk, OS, kernel, windowing system -- where before it listed
only Panama's own component versions. Graphics is joined from
GraphicsDevices rather than read again, because two readouts of the same
hardware are two things that can disagree. Placeholder DMI strings
("To Be Filled By O.E.M.") are filtered out, and unreadable facts are
omitted rather than shown as "Unknown".
The Fedora hand-off card was one row listing five subjects that opened
the network panel regardless. Naming a panel and then not opening it
reads as a broken button rather than a deliberate hand-off. Each subject
now opens the panel that owns it, and openGnomePanel takes an optional
subpage so "Users" reaches System's users page the way GNOME's own
desktop entry does, instead of dropping the user on System's front page.
Printers and online accounts are not repeated here; they stay with the
network hardware on Network & Devices.
One bug this surfaced, caught by the hyprland write contract: the Lua
config key and the hyprctl option name genuinely differ for tap to
click. hl.config wants input.touchpad.tap_to_click; getoption answers to
input:touchpad:tap-to-click. Either spelling used for both fails -- a
hyphen is not a Lua identifier, and the underscored name is not a known
option -- which is what the schema's two separate fields are for.
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
Continuing towards GNOME Settings parity. GNOME's Privacy panel covers
screen lock, camera and microphone access, file history, trash, and
device security; Panama had no equivalent page at all, despite already
tracking camera and microphone use for the bar indicator.
Device security is a new read-only readout: Secure Boot, TPM, disk
encryption, SELinux mode, and the firewall. None of these is a
preference -- they are set in firmware, at install time, or by system
policy, and a switch offering to change them would either fail or do
something far-reaching from a control that looks like every other
control. What it answers is "is this machine set up the way I think it
is", which otherwise takes five commands and root. Facts that cannot be
determined report Unknown rather than guessing, because a security
readout that quietly says "fine" when it failed to look is worse than
no readout.
File history and trash retention are deliberately NOT offered as
switches. They are GNOME preferences enforced by gsd-housekeeping, which
does not run in a Hyprland session -- verified, it is not running here.
Toggling them would store a preference, change nothing, and give no sign
of it. They are delegated to GNOME Settings by name instead.
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
Working towards parity with GNOME Settings, which splits pointing
devices into their own panel. Panama had pointer speed and focus-follows
buried under a page called "Input & Shortcuts", and had nothing at all
for scroll direction, acceleration profile, scroll speed, left-handed
buttons, or any touchpad setting -- all of which could only be changed
by editing hypr/input.lua by hand, which is the thing this app exists to
stop.
Every new mapping was read back off the running compositor rather than
assumed, and two were not what they look like: touchpad drag lock is an
int with three states, not a switch, and scroll factors are floats even
at their default of exactly 1. Getting either wrong makes every write to
that setting look rejected. The shape contract now covers 35 mapped
options, up from 23.
The touchpad card renders only when a touchpad is attached, which is
what InputDevices is for. On a desktop it would be worse than useless:
every switch on it would appear to work, because the preference is
stored and Hyprland accepts an option for a device class it has no
member of, so the settings would silently affect nothing.
Keyboard layout was read-only text, justified by a note saying changes
needed a compositor reload. That is not true in 0.56.2 -- setting
input:kb_variant through hl.config re-keymaps attached keyboards
immediately, verified by watching active_keymap on a real keyboard
change to "English (US, intl., with dead keys)" and back. So layout,
variant, and options are now real controls, joined by a TextEntryRow
that commits on Enter or focus loss rather than per keystroke, since
half a layout name is a valid string meaning something else.
Rejected input is shown as rejected rather than sanitised: these strings
are serialised into an hl.config payload, where stripping an unexpected
character would turn a typo into a different working setting.
Verified each new pointer option applies and reverts against the live
compositor. Schema, search, commit/reset, and system contracts pass.
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
brightnessctl drives the kernel backlight class, which laptop panels
have and this desktop does not -- it reports only keyboard and NIC LEDs.
So BrightnessControl removed itself and there was no way to dim the
screen from Panama at all. DDC/CI is the channel the buttons on a
monitor's bezel drive, and it is the only brightness an external display
has. Both sources now render a row each, so a machine gets whichever it
actually has, or none.
Displays are enumerated from sysfs rather than `ddcutil detect`. The
kernel publishes the connector-to-bus mapping as
/sys/class/drm/<card>-<connector>/ddc along with whether anything is
plugged in, which beats parsing detect's undocumented brief output,
yields the connector name spelled exactly as Hyprland spells it, and
probes only connectors with a monitor attached -- one bus on this
machine rather than fourteen, where each empty bus costs a timeout.
No model name is read: Hyprland already knows what every output is
called, so the UI joins on the connector instead of keeping a second
source of truth that could disagree with the Displays page.
Writes are debounced, serial, and read back. Serial because DDC/CI has
no arbitration and two ddcutil processes on one bus interleave their
exchanges and both return garbage. Read back because a write is not a
promise: panels clamp to their own range, ignore values while waking
from standby, and drop writes that arrive too fast. Without the read the
slider would show what Panama asked for rather than what the monitor
did, which is the same class of lie as trusting `hyprctl keyword`.
Brightness is deliberately not a stored preference. The monitor
remembers it and the bezel buttons change it behind Panama's back, so
persisting it would mean restoring a value the panel had moved past.
The contract runs against fixtures with ddcutil stubbed and both sysfs
roots redirected, so it never touches a real monitor. Its fixture
reports a maximum of 200 rather than 100 on purpose -- at 100 the
scaling arithmetic is the identity and a helper that ignored the
reported maximum would pass everything. Verified it catches that, plus a
dropped connection-status filter and an unstripped connector prefix.
Not yet confirmed against hardware: this machine cannot open any I2C bus
yet. ddcutil's udev rule grants that through uaccess but only to devices
created after it was installed, so it needs one udevadm trigger. The
helper detects exactly that case and returns the command as its error
rather than reporting "no displays".
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
The scheme switch already reached everything that reads
org.freedesktop.appearance -- GTK4, Qt6, Chromium, Electron -- because
ColorScheme.qml writes the gsettings key those all watch. Applications
carrying their own palettes did not follow, so choosing light mode left
the two windows actually used all day, kitty and neovim, still dark.
kitty: the 32 colours move out of kitty.conf into themes/, and kitty.conf
ends with `include current-theme.conf`. The generated file is machine
state rather than configuration, so it is gitignored and link-dotfiles
seeds it on install -- otherwise a fresh checkout starts by complaining
about a missing include. Running terminals are re-coloured in place over
their control sockets; a restart is not needed.
neovim: reads settings.json directly, since it neither watches the portal
nor keeps a socket open. Tokyo Night ships Day in the same family as
Moon, so light mode keeps the editor's identity instead of turning it
into a different-looking application. The existing readability overrides
were written against Moon and are now dark-only -- applied to Day they
would have put light grey on a light background, the same problem they
exist to fix, inverted. Light mode gets one override of its own:
tokyonight's shipped comment colour measures 2.54:1 against Day's
background, under the 3:1 floor for secondary text, so it is replaced
with 3.25:1 -- readable, still dimmer than Normal's 4.52:1.
An editor already open when the scheme flips re-applies on FocusGained,
which is cheap and fires exactly when the mismatch would be noticed.
Verified both directions: kitty re-coloured 4 live terminals, and neovim
starts as tokyonight-day with background=light and tokyonight-moon with
background=dark.
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
Light is Tokyo Night Day, the palette's own light variant, rather than
one invented to merely not be dark. Both share the same hues at
different lightness, which is what lets the Prism signature survive the
switch: blue still leads into orchid, it is simply a darker blue on a
lighter ground.
Every colour token became a binding on one boolean, so flipping it
repaints the whole shell without any component needing to know it
happened. That only worked because nothing outside Theme.qml defines a
colour; the two places that did are fixed here.
Surface alphas differ by scheme. The 0.34 that reads as glass over a
dark desktop reads as haze over a light one, and text stops being
legible on it.
Two things that draw on this desktop do not read Panama's store: GTK
applications, which read gsettings, and the compositor, which draws
window borders. A toolbar or a border still wearing the other scheme is
more jarring than either scheme on its own, so ColorScheme pushes the
choice to both. It also pushes at startup, since a scheme chosen in a
previous session would otherwise be in effect only for the shell.
The unfocused window border follows too. It is a flat neutral, and a
dark neutral is invisible against a light desktop. The focused border is
the prism gradient and needs no variant.
The live preview follows the scheme as well. A preview that stayed dark
while the shell around it went light did not read as "your desktop", it
read as a screenshot of someone else's.
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
Theme.qml was the largest remaining thing in this desktop that could
only be changed by editing a file, and typeface is the first thing
someone changes when they want a desktop to feel like theirs. Interface
font, icon font, and the base text size are settings now.
The two font choices are deliberately separate lists. Theme.fontMono is
used only to draw glyphs -- workspace pills, the status cluster, search
icons -- so a plain monospace family there replaces every icon in the
shell with tofu. The picker offers only Nerd Fonts for that slot and
says why.
Candidates are rendered in the family they name. A list of font names
set in the current font tells you nothing about what you are choosing.
The four type sizes derive from the base rather than being stored
separately, so the relationship between body, caption, heading and title
survives a change instead of four numbers drifting apart.
hypr/looks.lua reads the same key. Following the preference only on the
QML side would leave the compositor and the shell disagreeing about the
interface font, which nobody notices until a tooltip renders in a
different typeface.
Only a family this machine reports is accepted: the value reaches
hl.config as a string, and a settings file moved between machines will
name fonts that are not installed. A missing family is reported rather
than silently substituted by fontconfig.
Also collapses the wallpaper grid to two rows. Sixty tiles is two
screens of pictures on a page that also holds typography, window
geometry and effects -- everything below it was unreachable without
scrolling past all of them.
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
The last two values that could only be changed by editing a file.
Weather was pinned to hardcoded coordinates, so the card could not be
pointed anywhere else. It is a location search now, not latitude and
longitude fields: nobody knows their own coordinates, and a control that
demands them is one nobody uses. Open-Meteo's geocoding endpoint needs
no key, the same reason the forecast already uses them. Only the search
term leaves the machine -- the stored place name is a label -- and
coordinates are rounded to four decimals, far finer than a weather
reading resolves and coarse enough to keep a precise home location out
of the settings file.
The graphics readout was hardcoded to card1. This machine has two amdgpu
cards, discrete and integrated, so that was right only by luck, and the
path is meaningless on any other machine. GPUs are enumerated with a
readable name from lspci, since sysfs exposes only numeric ids, and the
picker appears only when there is more than one to choose between. A
stored path the machine does not have is refused and reported rather
than silently measuring nothing.
Also merges the per-application notification rules UI. Its three commits
were believed integrated but the page half was not actually in the tree:
main had the service side in Notifs.qml and zero references to
setAppRule in NotificationsPage. Ancestry is not content.
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
Audited bar, dock, quick settings, date menu and Settings for three
things: a setting reachable in one UI but not another, a setting that
exists but is unreachable anywhere, and UI that states something false.
Night Light was fully exposed in Quick Settings and had no control
anywhere in Settings. It now has a card on Displays, where GNOME also
puts it, with on/off, schedule, times and temperature.
Adding those controls would have shipped the exact defect this audit
exists to find. NightLight declared enabled, temperature and automatic
as bindings on the store, but toggle() assigns to them, and an
assignment destroys a QML binding permanently -- so the service wrote to
the store and never read from it again. The Settings controls would have
written values the service ignored, while Quick Settings kept working.
It now follows the store. Every other service was swept for the same
pattern; this was the only one.
The night light schedule was two hardcoded literals, so the hours could
not be changed. They are schema keys now, with a row that renders 17.5
as "5:30 PM" and honours the 24-hour preference rather than showing a
decimal nobody reads as a time.
keyboardLayout was in the schema and read by input.lua but had no
control anywhere: configurable in principle, unreachable in practice. It
is surfaced on Input & Shortcuts as read-only, with the reason, because
it needs a compositor reload and a control implying instant apply would
be a smaller lie but still a lie.
Caffeine was a Quick Settings toggle mentioned only in a subtitle in
Settings. It has a real control now.
Claude-Session: https://claude.ai/code/session_01BRvzt4H8XXLPVH5MyYdk9L
Gabriel Brown