Run 31394740824 compiled all 27 components on all three distributions --
two and a half hours each -- staged them, passed the completeness
assertions, and then failed in the packaging step on every one. Three
separate causes, none of them the desktop.
desktop-file-validate rejects DesktopNames (fedora, arch)
hyprcosmic.desktop: error: file contains key "DesktopNames" in group
"Desktop Entry", but keys extending the format should start with "X-"
DesktopNames is the key a display manager reads to set
XDG_CURRENT_DESKTOP, so a session entry needs it, but the Desktop Entry
Specification registers keys for application launchers and the validator
has no entry for it. This is not something the fork introduced:
cosmic.desktop is upstream cosmic-session's file, unchanged apart from the
Exec path, and the copy already installed on this machine as
cosmic-session-1.5.0-1.fc44 fails the identical check. Dropping the key
would satisfy the validator and break the session.
So the validation moves out of the three packaging recipes into one step
against the staged tree, and checks what desktop-file-validate would not
have caught anyway: that Exec is absolute and names an executable this
package actually installs. A wrong Exec puts an entry on the greeter's
menu that dies silently when it is chosen, which is the failure that
matters here; a key the validator has not heard of is not.
The check was run against a staged tree built from the real desktop files
before committing: it passes clean, and it rejects an Exec naming a binary
that is not in the tree, an Exec that exists but is not executable, a
relative Exec, a missing DesktopNames, a missing Name, a missing [Desktop
Entry] header and Type=Link, each with a message saying which file and
what.
SIGPIPE under pipefail (debian, and arch next)
dpkg-deb: error: tar subprocess was killed by signal (Broken pipe)
The Debian package built: 217 MB, correct control, Depends computed by
dpkg-shlibdeps across 17 libraries. It failed on `dpkg-deb --contents
dist/*.deb | head -20` in the verify step -- head closes the pipe after
twenty lines, dpkg-deb dies of SIGPIPE, and Actions runs these steps with
pipefail. Both now write to a file and head the file. The Arch verify step
had the same line waiting for it and is fixed in the same way.
bogus date in %changelog (fedora)
10 August 2026 is a Monday.
Also: the PKGBUILD still pointed at tools/make-packages.sh, which was
deleted when packaging moved to Actions. It now names the two commands
that stage a tree and build from it by hand.
Run 31383454158, arch job. It compiled all 26 other components over two
hours and eighteen minutes and then failed on the last line of `build`:
/usr/sbin/make -C xdg-desktop-portal-cosmic all
make: *** No rule to make target 'all'. Stop.
xdg-desktop-portal-cosmic has no Makefile. It has a justfile, with `build`
and `install` recipes, and the submodule is clean at the commit this fork
pins.
This is upstream's, not something introduced here. The line dates from
2022 (f4fb6a86), pop-os/cosmic-epoch master still carries it verbatim, and
upstream pins the same portal commit this fork does -- f211aa37,
epoch-1.5.0 -- so `just build` on upstream cosmic-epoch fails in exactly
the same place. The portal moved from a Makefile to a justfile and the
meta-repository's recipe was never updated. It goes unnoticed because
distributions build COSMIC one component at a time and never take this
path; it takes a full meta-build to reach the failure, and the failure is
on the very last line.
`build` rather than `build-release`: the portal has no build-release
recipe, and its `build` defaults to debug='0', which selects --release.
`install` takes rootdir/prefix rather than DESTDIR/prefix, being a
justfile.
Every other delegation in this justfile was checked the same way -- each
`make -C dir target` against that directory's Makefile targets, each
`just dir/recipe` against that directory's recipes, following justfile
imports -- and the portal was the only one that did not resolve.
Run 31383454158, debian job, eighteen minutes in:
error: failed to run custom build command for `dav1d-sys v0.8.3`
Package 'dav1d', required by 'virtual:world', not found
dav1d is pulled in by cosmic-bg, and nothing in the obvious set of
compositor dependencies names it. A hand-assembled list will keep having
that shape of hole, and each one costs a full run to find.
So the lists are now the union of Build-Depends across all 27 upstream
components' debian/control files, collected mechanically. That adds
libdav1d-dev, libglib2.0-dev, libspa-0.2-dev, libegl1-mesa-dev,
imagemagick, intltool and fonts-open-sans, none of which were present
before and all of which upstream declares.
The Fedora and Arch translations were verified against `dnf repoquery` and
archlinux.org's package API rather than guessed, because a name that does
not exist fails the whole step and costs another run to discover: it is
libdav1d-devel on Fedora and not dav1d-devel, and ttf-opensans on Arch and
not otf-opensans. Both mistakes were in the first draft of this commit.
Run 31383454158, fedora job, first line of the build:
justfile:3: make := `which make`
sh: line 1: which: command not found
Fedora's container image is deliberately minimal and has neither `which`
-- deprecated there in favour of `command -v` -- nor make. Arch's
base-devel and Debian's build-essential include both, which is why this
broke on exactly one of the three.
Rather than fix that one symptom and rediscover the next, the recipes were
read for what they actually shell out to: 360 cargo, 163 install, 114 tar,
26 which, 18 find, 11 sed, 4 xargs, 2 make. findutils and tar are also
absent from Fedora's image, so both would have failed in turn, one run at
a time. All of them are now named, along with gzip, sed and diffutils.
Debian gains `file`, used by the deb step to pick ELF binaries out of the
staged tree for dpkg-shlibdeps. It is not part of Debian's essential set.
rust-cache is removed rather than repaired. It runs `cargo metadata` at the
workspace root to decide what to cache, and this root is a meta-repository
with no Cargo.toml -- 29 submodules, each its own crate -- so it reported
"could not find Cargo.toml" and cached nothing. Listing all 29 workspaces
would fix the error and buy a worse problem: their target directories come
to roughly 17 GB against a 10 GB per-repository limit, so the three jobs
would evict each other every run and pay the upload time for it. A cold
build is the honest cost of a workflow that runs on tags and on demand.
Both failures on run 31383204572, all three distributions, in the same
step, before anything was compiled.
--default-toolchain none was copied from the two fork workflows, where it
is correct: their checkout root is the crate, so rust-toolchain.toml sits
in it and rustup reads the pin. This repository's root has no pin -- it is
in cosmic-comp/, one level down, and the other 27 components have none at
all -- so `none` left no default and the first cargo invocation at the
root failed with "could not choose a version of cargo to run".
stable does not weaken the pin. rustup applies a directory-local
rust-toolchain.toml on entering that directory and installs it on demand,
so cosmic-comp still gets the 1.93 it asks for.
The same logs showed `shell: sh -e {0}`, which is the container default
and is dash on Debian. Two later steps use brace expansion, which dash
lacks and bash-as-sh disables, so the RPM build would have created a
directory literally named rpmbuild/{BUILD,RPMS,...} and failed further
along. Declared bash for the job rather than rewriting around a constraint
none of the three images impose.
Worth recording why this was not caught before pushing: the run blocks
were syntax-checked with bash, which is not what was going to run them.
Checking with dash would not have caught it either -- brace expansion
failing is runtime behaviour, not a parse error. Declaring the shell is
the fix; there is no static check that substitutes for it.
Bumps both forks to the commits that install at upstream's paths, and adds
the packaging that follows from it.
packaging/ holds one definition per distribution -- an RPM spec, a PKGBUILD
and a Debian control template -- and each wraps a tree that `just install`
has already staged rather than compiling again inside the packaging tool.
Building 27 Rust components a second time to produce bytes that already
exist costs hours and creates a way for the packaged desktop and the built
one to drift apart.
One package per distribution, not one per component. Fedora splits COSMIC
into 27 packages, which is right for a distribution tracking upstream. This
is a fork that replaces the desktop as a unit: there is no supported
combination in which you take the HyprCosmic cosmic-comp and the
distribution's cosmic-session, and one package says so accurately.
All three declare a conflict with the distribution's cosmic-comp and
cosmic-session, and stop rather than resolve it. Obsoletes would let a
routine install quietly remove the desktop the machine is currently
running; removing COSMIC stays a decision a person makes.
.github/workflows/packages.yml builds all three, each inside a container of
the distribution it targets, because nothing here is statically linked and
a package built elsewhere records sonames the installing machine will not
have. It runs on tags and on demand rather than on every push -- three full
desktop builds is hours of runner time for artifacts nobody downloads --
and a tag opens a draft release, not a published one.
The RPM file list is generated from the staged tree, and claims a directory
only when no package already owns it. A naive list would have the package
own /usr, /usr/bin and /usr/share, which belong to `filesystem`.
README follows the same reframe: the install goes to /usr/bin, the package
route is documented first, and the conflict is explained where a reader
meets it rather than left to be discovered.
The header said the binaries were out of scope because `cargo` installs them.
That was true when they were built by hand in three separate checkouts. The
top-level justfile now installs all three, so the sentence pointed at the wrong
thing -- someone looking for where cosmic-conf lands would have gone to cargo
and found nothing. The scope is unchanged; only the reason was stale.
Both files said `name: CI`. The concurrency group is built from
${{ github.workflow }}, which is that name, so the two workflows shared a group
and cancelled each other on every push -- whichever started second would be the
only one that ever reported a result. The compositor and session forks already
say HyprCosmic for the same reason; this one was missed when it was renamed.
The tree still carried cosmic-epoch's README, which describes how to install
COSMIC from a distribution's packages -- accurate, and about a different thing
than the repository it is now sitting in. Anyone arriving here needs to know
what changed, which two submodules are ours, how to build it, and what to do on
first login, and none of that was written down anywhere outside commit messages
and the comments in config/cosmic.conf.
Deliberately not a copy of upstream's. The dependency list, the packaging notes,
the per-distribution install instructions and the translation links are all
still correct and all still upstream's, so this links to them rather than
forking a second copy to go stale. What it says instead is the delta: the four
cosmic-comp patches, the session profile and what it stops starting, the
one-way config model, the theme import, the four per-user files `just install`
deliberately does not place, and what the two CI workflows each cover.
Two things in it are warnings rather than instructions, and are there because
leaving them out would make the document a nicer read and a worse one:
- prefix must be /usr in practice, and the reason is that some files name
/usr/share/hyprcosmic literally because they have no way to interpolate a
prefix. Written as its own paragraph rather than a footnote.
- `install` depends on `build`, so `sudo just install` compiles as root. That
is upstream's arrangement, inherited rather than chosen, and says so.
Every command, path and claim in it was checked against the tree: the CLI help,
the profile's disabled list, the socket names, the power menu's five entries,
the Shortcuts merge, and the theme path HyDE's own generated config records.
`just build` built 26 upstream components and our two forks. `just install`
installed the same set. Neither touched cosmic-conf or anything under config/,
which meant the recipe that is supposed to produce a desktop produced one with
no config compiler, no bar layout, no rofi theme and no power menu -- a
HyprCosmic session that comes up to a blank screen and cannot be logged out of.
So three lines, in the places the existing recipes already establish:
build cargo build --release --manifest-path cosmic-conf/Cargo.toml
install the binary to $prefix/bin, then tools/install-assets.sh
clean cosmic-conf/target
cargo directly rather than `just cosmic-conf/build-release`, because cosmic-conf
is a crate in this repository rather than a submodule and has no Justfile of its
own to delegate to.
install-assets.sh runs last and with --no-session. Last because it is the only
step whose output is worth reading: several of the files it places name
/usr/share/hyprcosmic as a literal -- a .rasi has no variables and the autostart
file is deliberately not a shell -- so a prefix other than /usr installs them
where nothing will look, and the script says which ones. --no-session because
cosmic-session/install has already placed start-hyprcosmic and
hyprcosmic.desktop by then, and installing them twice would leave it unclear
which recipe owns them.
Verified by running the two new install lines verbatim into a staging root:
eight files, the cosmic-conf binary and the seven shared assets, and no session
entry point among them.
Two consequences of becoming the meta-repo, neither of which the merge itself
could have caught.
The submodule URLs were written as [email protected]: because that is how they are
pushed from here. That is a working copy's business, not the repository's:
upstream names all 29 over HTTPS, a clone with no key configured is the normal
case, and CI has no key at all. The URLs are now HTTPS; each submodule's own
origin stays SSH, which is where it belongs.
The assets job asked for `submodules: recursive` back when this repository had
two of them. It now names 29, and 27 are the rest of COSMIC -- several gigabytes
fetched per run to check that some CSS is in step with its generator. It fetches
the one it needs instead.
That one is cosmic-session, because install-assets.sh also places the session
entry point that lives in that fork. The script skips those two files with a
warning when the checkout is absent rather than failing, so a fetch that
silently stopped working would quietly reduce what the job covers instead of
turning it red. Hence the `test -f` after it.
HyprCosmic started as a repository beside COSMIC holding a config compiler and a
pile of theme assets, with the two modified components hanging off it. That had
the relationship backwards. COSMIC already has a meta-repo whose whole job is to
name every component and build the desktop from them, so the honest shape for a
fork is to be that meta-repo with the components we changed pointing at our
copies -- not a separate tree that assumes the rest of COSMIC arrived some other
way.
So this merges cosmic-epoch in and repoints exactly two of its 29 submodules:
cosmic-comp -> outbackdingo/hyprcosmic-comp
cosmic-session -> outbackdingo/hyprcosmic-session
The other 27 stay on pop-os. Nothing about them needs to change, and pinning
them to copies we do not maintain would be a promise to keep 27 forks current.
`just build` and `just install` now build and install the whole desktop with our
compositor and session in it, which is what "fork COSMIC" ought to mean. The two
forks still install into /usr/libexec/hyprcosmic rather than over /usr/bin, so a
system that already has COSMIC from its distribution keeps that session on the
greeter's menu alongside this one.
Conflict resolutions worth stating:
.github/workflows/ci.yml upstream's kept as-is. It builds the entire desktop
on Arch through `just sysext`, which is precisely
the check a meta-repo wants and is not made less
useful by forking. Our per-distribution workflow
moved to hyprcosmic.yml beside it, the same way it
did in the compositor fork.
.gitmodules upstream's 29 entries, then two URLs rewritten.
cosmic-comp, cosmic-session
ours. Git reports "no merge base" because they are
unrelated to the pop-os commits recorded here,
which is expected: they are different repositories,
not newer commits of the same one.
.gitignore both sides, plus a note that the two forks are
deliberately no longer ignored.
They were in .gitignore with a note saying they would become submodules once
they had somewhere to live. They do now, so they are, and ignoring them would
from here on only hide which commit of each one this tree is built against.
The paths stay cosmic-comp/ and cosmic-session/, which is where cosmic-epoch
puts them, so this repository can be merged with the upstream meta-repo rather
than laid out beside it.
Two lines named /home/dingo outright: waybar's stylesheet and the wallpaper the
`current` symlink points at. This file is installed verbatim into every user's
config directory, so both worked for exactly one person -- and failed quietly
for everyone else, because a waybar whose stylesheet cannot be read still starts
and a wallpaper that was never set looks the same as one that failed to load.
Nobody would have got an error message; they would have got an unstyled bar over
a black screen.
The wallpaper line was already a shell invocation, so it only needed $HOME.
The bar was not, and this file is deliberately not a shell -- `~` and `$HOME` on
a bare command line here are literal text, which is what stops a file that names
programs from being escalated into arbitrary execution. So it now goes through
`sh -c`, on the same terms the wallpaper line already established: naming `sh`
is naming a program, and anyone who can write this file could already name any
binary on the system. `exec` keeps the process tree flat, which matters because
cosmic-session supervises these -- without it the shell would be what gets
supervised, and a waybar that died would never be restarted.
Verified that the expansion produces the same two paths that were hardcoded, and
that both exist.
Two jobs, shaped differently on purpose.
cosmic-conf gets the same three-distribution container matrix the forks use,
plus one check the unit tests cannot make: it resolves the cosmic.conf this
repository actually ships. A schema change that invalidated the shipped config
would pass all 143 tests in the crate and still break every user on their first
login. The binary is invoked directly rather than through `cargo run`, because
the step redirects HOME and cargo keys its registry cache on it -- `cargo run`
would re-download every dependency into a directory the cache action does not
know about.
The assets job is where the rules that were previously only remembered become
enforced. config.jsonc is regenerated and must not move, which is what stops it
being hand-edited; the template and the generator are held to pure ASCII, which
is what stops a Nerd Font glyph being pasted somewhere it will be silently
destroyed by the next person who retypes it. install-assets.sh is then run into
a staging root and asked to verify its own work, which also exercises the audit
that refuses to install at all while any file under config/ is unclassified.
The ASCII check is written as an `if` rather than `! grep`, because grep exits 1
for "no match" and 2 for "no such file" -- negating it would turn a vanished
file into a pass, and the check would quietly stop checking anything.
Focus follows mouse was already there and already off. cosmic-comp has
supported it for as long as this fork has existed -- focus_follows_cursor and
focus_follows_cursor_delay, both live-watched -- and the schema already exposed
them under general. What was missing was the Hyprland spelling: there was no
input section at all, so a config written the way a Hyprland user would write
it named a setting that did not exist.
So this is an alias, not a new setting. input.follow_mouse and
input.follow_mouse_delay resolve to exactly the same cosmic-config keys as
general.focus_follows_cursor and its delay, and both spellings stay. Setting
both in one file is not an error; the last assignment wins, which is the rule
the rest of the file already follows. A test pins the two pairs to the same
targets, because a rebase that renames a target key would otherwise leave one
spelling working and the other quietly dead.
follow_mouse needs its own type. Hyprland writes it as a number and COSMIC
stores a bool, so Ty::FollowMouse maps 0 to false and 1 to true -- the same
trick Ty::Mode already plays for dark/light. Hyprland's 2 and 3 split pointer
focus from keyboard focus, which cosmic-comp cannot express: it has one focus
and either moves it or does not. They are rejected with a diagnostic that says
why, rather than rounded up to 1, because silently handing click-to-focus to
someone who asked for the opposite is worse than telling them the mode does not
exist here. Anything else gets the ordinary "expected 0 or 1" error.
There is no autoraise key because autoraise is not a separate feature.
raise_with_children runs inside update_active, which is what the
focus-follows-cursor timer ends up calling, so a floating window under the
pointer comes to the front as part of being focused. Tiled windows do not
overlap, so raising one is a no-op.
The delay is left at COSMIC's 250ms rather than shortened. It is what stops
focus from skating across every window the pointer crosses on its way
somewhere else, and that failure is more irritating than the wait.
Installing the binary before editing the config is the required order, not a
preference: resolution is transactional, so the old binary meeting an unknown
input section would refuse to write the whole file, not just that block.
The hyprcosmic profile disables cosmic-panel, and COSMIC's power applet lives
in that panel. Nothing replaced it, so the session had no logout, reboot or
shutdown anywhere in it: the only way out was `systemctl reboot` typed into a
terminal, which also meant every login-time change cost a reboot to test.
hyprcosmic-powermenu is a rofi menu offering lock, suspend, log out, reboot and
shut down. It is reached two ways -- $mainMod SHIFT E and a button at the right
end of waybar -- and both run the same script, so a click cannot bypass the
confirmation a keypress gets.
Logging out calls com.system76.CosmicSession.Exit, which is the method the
panel applet used and the only one that stops the session's clients in order
rather than pulling the compositor out from under them. Reboot and poweroff go
straight to systemd; polkit already authorises both for an active local session
without a prompt, verified with pkcheck, so no pkexec is involved. Lock uses
loginctl, which cosmic-greeter is listening for.
Confirmation is asked only for the three that end the session. Lock and suspend
undo themselves with a keypress, so a prompt there is pure friction; the other
three throw away everything unsaved and are one keystroke away at all times.
"No" is listed first so it is the row already selected.
The menu entries are plain words rather than Nerd Font glyphs. The bar icon is
the only glyph involved, and it comes through generate-config.py, whose whole
purpose is that no Private Use Area character is ever typed by hand -- U+F011,
confirmed present in the installed JetBrainsMono Nerd Font.
rules.css names every module id explicitly, so #custom-power had to be added
there too or the button would have rendered with no pill behind it.
The script sits under config/bin/ rather than a new top-level directory so that
install-assets.sh's audit still covers it: that check refuses to run unless
every file under config/ is classified, which is what stops a new file from
being silently left uninstalled.
`general.preserve_split` maps straight to the compositor key Patch C adds,
so opening a third window gives three panes in a row instead of quadrants.
Every other key under `general` exists upstream; this one is the fork's
own field on `CosmicCompConfig`. If a rebase ever loses that field the
compositor would ignore the key without saying anything, so the target is
pinned in a test of its own rather than left to the generic registry
checks.
COSMIC ships autotile off, so every new window opened floating at whatever
size the application asked for, on top of whatever you were looking at. The
key already existed in cosmic-conf's registry and had simply never been set.
Gaps come with it. They are invisible until windows tile -- with nothing being
laid out, nothing has a gap -- so setting one without the other is half a
change. gaps_out is doubled so the screen edge reads as margin rather than as
one more seam.
autotile_behavior is deliberately not a conf key. It defaults to Global, which
retiles workspaces that already exist rather than only arming new ones, and
that is the value worth having; anyone who wants PerWorkspace can set it in
cosmic-settings without this file overwriting them.
tools/verify-hypr-ipc.py exercises the fork's Hyprland IPC the way a client
reaches it, which the compositor's own log cannot show you. It checks the
socket names, that the five read commands answer JSON, that `bogus`,
`dispatch exec rofi`, `dispatch killactive`, `dispatch workspace +1` and
`dispatch workspace 0` are all refused, and that a real switch returns `ok`.
It moves the focused workspace, so it returns to the one you started on.
It is a validated negative test, not a hopeful one: run against the old
compositor before the socket rename was installed, it failed on the names and
exited 1.
docs/unreproducible-dead-input-2026-08-10.md records the session that came up
with no keyboard or pointer at all, and did not come back after a reboot. It
is closed deliberately rather than fixed, and most of its value is the list of
things it is not -- the fork's patches, `seats.for_device()` returning None,
the modifier-only Super binding, a shortcuts-config race, and two scary log
lines that stock COSMIC prints too. The one suspicious fact is that it was the
fourth compositor start on that boot. If it recurs, there is a list of what to
collect before rebooting destroys it.
The design spec said `.socket` and `.socket2`; corrected to the names clients
actually open.
cosmic.conf's bare-Super binding gets a comment saying why the key field is
empty, since an empty field in a `bind` line reads like a typo. COSMIC
supports modifier-only bindings and Hyprland's `bind` cannot express one.
The bar had six modules and one dead click: pulseaudio's on-click ran
pavucontrol, which is not installed. It now opens `cosmic-settings sound`,
with middle-click as a mute toggle. Added: mpris, bluetooth, temperature,
idle_inhibitor, privacy, a swaync notification button, and power-profiles-
daemon -- and HyDE's pill styling, so each module is its own rounded chip
rather than text in a row.
No backlight module. This machine's panel has no sysfs backlight interface, so
it would render as a permanent error.
The rest of this is about the icons, which have now been got wrong enough
times to deserve a mechanism.
Private Use Area characters do not survive being typed. Writing the previous
version I put a comment at the top of the file saying every glyph was an
escape, then typed literal glyphs into the same file; the codepoint dump found
`format-bluetooth` had picked up a stray U+F293 and another field had two
glyphs where I had written one. Nothing errors -- waybar is perfectly happy to
render a label that is one space.
So the config is generated, not written. config.jsonc.in is pure ASCII with
@@TOKEN@@ placeholders, generate-config.py holds the name-to-codepoint table
and emits `\uXXXX` escapes, and it asserts its own output `.isascii()` before
writing. A hand-typed glyph now cannot reach the file. The delimiter is
doubled because single `@NAME@` collided with wpctl's `@DEFAULT_AUDIO_SINK@`,
which the generator caught as an unknown token rather than mangling.
Every codepoint was checked against the installed font with `fc-list
":charset=..."`. HyDE's own values do not all survive that: its muted glyph
U+FA80 is an old Material Design Icons codepoint that Nerd Fonts v3 moved, so
copying upstream verbatim would have shipped tofu. Three were replaced.
The font stack named "FontAwesome 6 Free" first, which resolves to the Regular
face and carries a fraction of the icon set -- the icons that did appear were
coming from accidental per-character fontconfig fallback. JetBrainsMono Nerd
Font goes first now.
One CSS note, because the failure mode is not obvious: GTK has no `:empty`
pseudo-class, and an unknown pseudo-class does not skip the rule, it rejects
the whole stylesheet and waybar exits 1. `#tray:empty` took the bar down.
install-assets.sh grows a third category. The template and the generator live
under config/ but must not be installed -- a file full of placeholders sitting
next to the real config is a coin toss for whoever opens one first -- and the
audit refuses to run until every file is classified, which is exactly what it
is for.
Everything outside $HOME was placed with `sudo install` while the desktop was
being built, which left two problems.
A fresh machine has none of it, and the failure is loud in the worst way:
config.rasi imports /usr/share/hyprcosmic/rofi/{palette,rules}.rasi by absolute
path, and a missing @import is an error rofi renders *in place of the
launcher*, not a warning it skips. Miss those two files and Super+A shows a
parse error.
And hand-installed files drift. Writing this found that /usr/bin/start-
hyprcosmic had silently gained a session-logging block during the blank-screen
debugging that never made it back to the copy under version control -- found
by diffing the two on a hunch, which is not a strategy. `--check` compares
every managed file against its source and exits non-zero on any difference.
PER_USER is not documentation. The script refuses to run unless every file
under config/ is listed as either shared or per-user, so adding one forces a
decision about which it is instead of letting it be quietly left out of both.
PREFIX is only half honoured and the script says so rather than pretending
otherwise: rofi's .rasi has no variables and the autostart file is explicitly
not a shell, so both name /usr/share/hyprcosmic literally. The warning finds
them by grep rather than from a hardcoded list, so it cannot go stale -- it
already turned up waybar/style.css, which I had not thought of.
No internal sudo. It probes the nearest existing ancestor of each destination
up front and dies with the exact command to re-run, rather than escalating on
its own or failing half way through.
The session entry point comes from the cosmic-session fork, which is a
separate checkout and may be absent; it is skipped with a note when it is, or
by --no-session. Binaries are out of scope: they are build outputs, so
comparing them byte-for-byte would only ever report a rebuild.
Verified all four modes against a DESTDIR staging tree and the live /usr:
drift and missing files are detected and exit 1, a re-install repairs them, an
unclassified file under config/ is refused, a bad argument is refused, modes
land as 644 and 755, and `--check` against /usr now reports all 8 files
matching.
The line named a wallpaper inside the theme directory directly, so importing a
different theme left it pointing at a path that no longer existed -- a blank
screen at the next login, with nothing in the log to say why, because nobody
asked for a wallpaper and so nothing reported one missing.
It now names ~/.local/share/wallpapers/hyprcosmic/current, the symlink
`import-theme --assets` maintains. rofi's local.rasi shows the same image in
the launcher sidebar and names the same link, so the two cannot drift.
Changing the wallpaper is now `ln -sfn`, not an edit to this file, which the
comment says so that the next person does not undo the indirection.
The path stays literal. This file is not a shell -- deliberately, so that a
file naming programs cannot be escalated into arbitrary execution -- so `~`
and `$HOME` would be passed through as text.
The four-layer rofi chain needs config.rasi and local.rasi to exist, and
neither has an upstream file to copy: a HyDE theme has no equivalent of
either. They were hand-written for this machine, which meant a second machine
got a launcher that reported a missing @import instead of opening.
config.rasi is `include_str!`d from config/rofi/ rather than kept as a string
literal, so it stays a real .rasi file -- highlightable, diffable, editable
without a rebuild to see the result -- and there is one copy of it rather than
two that can disagree.
local.rasi is composed per machine from two things the repo cannot know: the
theme's $ICON_THEME, and a wallpaper path. Each half is omitted entirely when
there is nothing to say, because an empty `icon-theme:` list is something rofi
would honour.
$ICON_THEME comes back as a field on Import rather than being re-parsed out of
the conf text that the same function just rendered.
The wallpaper needed a stable name. local.rasi cannot hardcode a filename
without going stale at the next theme import, so plan_wallpapers now also
maintains ~/.local/share/wallpapers/hyprcosmic/current as a symlink to one of
the copies it made. HyDE has this problem too and solves it the same way, with
~/.cache/hyde/wall.thmb. The launcher sidebar and the autostart's `awww img`
line both name the link, so they cannot drift apart.
Which wallpaper it points at is the first in *sorted* order. read_dir returns
whatever the filesystem feels like, and an arbitrary choice is fine where an
unrepeatable one is not: re-running the import would otherwise change the
wallpaper at random. The link is repointed even when every wallpaper was
skipped as already installed -- the copies are theme-specific and unchanged,
but the link is global and has to follow the theme just imported.
Repointing has to handle a *dangling* link, which is exactly what a previous
import leaves behind once its theme directory is gone: symlink(2) fails with
EEXIST rather than replacing, and Path::exists follows the link, so it answers
false for the one case that needs removing. symlink_metadata asks about the
link itself.
Values reaching a generated config come from a theme directory that may have
been downloaded from anywhere, so quote_rasi_string strips quotes, backslashes
and control characters -- .rasi has no escape syntax worth relying on.
Four existing tests counted actions and broke once every plan carried two more.
Fixed with a theme_assets() filter rather than by bumping the numbers, so what
they are actually asserting stays visible and the next generated file does not
break them again.
125 unit, 5 bin and 4 integration tests pass; clippy --all-targets and
`cargo fmt --check` are clean. A real import of Tokyo Night installed all 13
paths, and `cosmic-conf apply --diff` reports no changes afterwards.
`rofi -show drun` came up unstyled. Three separate reasons, all of them
invisible: `import-theme` wrote ~/.config/rofi/theme.rasi, but rofi only ever
auto-loads config.rasi and nothing imported theme.rasi; theme.rasi holds a
HyDE palette and nothing else, no widget geometry at all; and no layout was
shipped anywhere for it to colour.
The layering mirrors what config/waybar already does, for the same reason --
a theme supplies colours, and the layout has to survive being handed a theme
that defines only some of them:
1. palette.rasi defaults, under the exact names HyDE themes use
2. theme.rasi the installed theme's rofi.theme, written by cosmic-conf
3. rules.rasi geometry and layout, with no colour literals at all
4. local.rasi per-machine paths: sidebar wallpaper, icon theme
Later imports win, so a theme recolours the launcher without rules.rasi
knowing a theme exists, and a machine points at its own wallpaper without
either of them knowing the path. Unlike waybar there is no bridge step: a HyDE
rofi.theme defines the names rules.rasi already references.
1 and 3 are shared and go under /usr/share/hyprcosmic/rofi. 2 and 4 must be
per-user, and config.rasi with them: it imports those two relatively, and rofi
resolves a relative @import against the importing file's directory.
rules.rasi is HyDE's style_1 with three deliberate departures, documented in
its header. HyDE computes the border width, radius and font in rofilaunch.sh
before invoking rofi; there is no launcher script here -- the keybinding runs
`rofi -show drun` bare -- so those are baked in. The sidebar image moves to
local.rasi because HyDE's ~/.cache/hyde/wall.thmb does not exist outside HyDE.
And dummywall gets `background-color: @main-bg` rather than transparent, so a
machine with no wallpaper set shows a panel instead of a hole.
The display-* labels are Nerd Font glyphs copied byte-for-byte out of
style_1.rasi. They are Private Use Area codepoints and do not survive being
retyped; verified as U+F303, U+F120, U+F07B, U+F2D0, each followed by a thin
space, and confirmed present in JetBrainsMono Nerd Font via
`fc-list :charset=`.
Verified with `rofi -dump-theme` and `-dump-config`: both exit 0 with empty
stderr, and the merged dump shows the Tokyo Night theme's main-bg beating the
palette default.
Without it "the file wins" only held at the moment someone last ran `apply` by
hand. Now cosmic.conf is compiled at login and recompiled on every edit to it
or to anything it sources, so whatever COSMIC's settings UI has stored since
the last login is overwritten before the desktop settles.
First in the file for that reason. The bar does not read cosmic-config, so the
ordering is for the compositor's benefit rather than waybar's.
No `--config`: the default path is derived from XDG_CONFIG_HOME inside the
process, so unlike the waybar line this needs no hardcoded home directory --
the one part of this file that is not portable as written.
A malformed edit stays non-fatal. It goes to the session log and the last good
configuration remains in place, so a typo cannot strand you at a broken
desktop; fix the file and the next save applies.
Requires cosmic-conf on PATH, which it now is (/usr/bin/cosmic-conf, release
build). `waybar` and `awww-daemon` are already named bare here, so the profile
resolves argv[0] through PATH.
`watch::watch` has been written, tested and unreachable from the CLI since it
landed. It now has a command: `cosmic-conf watch [--config <path>]`, sharing
`--config` with `apply` and refusing `--diff`, which means nothing for a
daemon whose whole job is to notice a change and write it.
Exposing it made an existing wart user-visible: a single bad save reported
itself three or four times. One write arrives as several inotify events --
modify, close_write, and a rename when the editor writes atomically -- and
they do not all land inside one 250ms debounce window, so each produced its
own compile and its own copy of the same diagnostic. Consecutive identical
errors are now printed once, reset on any successful compile so the same
error after a good one is still news.
Verified against an isolated XDG_CONFIG_HOME, driving a real daemon rather
than calling `compile` directly, since none of this is reachable from the unit
tests: applies at startup, recompiles on edit, notices edits to sourced files,
picks up a `source` line added at runtime, survives a malformed edit with the
last good value intact, reports it exactly once, does not suppress a
*different* error, and resumes after a fix. Ten checks, all passing.
Mostly mechanical -- writeln! for format strings ending in a newline,
sort_by_key, slice::from_ref, &Path over &PathBuf, a stray &mut in a test.
Two were worth more than the lint that found them.
`parse_color` sliced `hex[i..i + 2]` after checking `hex.len()`. Both are byte
counts, so a multi-byte character inside `rgb(...)` split a char boundary and
panicked: `rgb(€abc)` is six bytes and aborted the compiler with "end byte
index 2 is not a char boundary". A typo in a config file must produce a
diagnostic, not a crash. Clippy did not see this -- it flagged the duplicated
`rgb(`/`rgba(` arms as foldable into `?`, and folding them is what put the two
length assumptions next to each other where the mismatch was visible. Now
guarded by is_ascii, with a test that panics without the guard.
`plan_verbatim` tripped too_many_arguments at 8. Three of them were the
`&mut Vec<Action>`, `&mut Vec<Note>` and `&mut Vec<AssetError>` threaded
through both plan helpers -- a Plan under construction, so `Draft` now names
it and `finish()` owns the errors-are-fatal rule that was previously inline.
118 unit tests plus 4 integration tests pass; `cargo fmt --check` and
`cargo clippy --all-targets` are both clean.
rustfmt and clippy were installed all along; I had wrongly recorded them as
missing and never ran either. This is the mechanical half -- files rustfmt
reformatted and nothing else touched, committed separately so the real fixes
in the next commit are readable.
`cosmic-conf apply --diff-only ~/.config/hyprcosmic/cosmic.conf` did exactly
the wrong thing twice over: the misspelt flag was ignored, so it wrote instead
of diffing, and the path was ignored too, so it wrote to whatever the *default*
config compiles to. It reported success either way.
Both arguments are now errors with exit 2. An argument parser that silently
skips the unknown is a bad fit for a command whose job is to overwrite
settings; the surface here is six flags, so the check is fifteen lines rather
than a dependency.
Keeping the imported theme in its own file, sourced from cosmic.conf, is what
stops a re-import from clobbering the keybindings. But a sourced file only
does anything if something sources it, and until now `import-theme --out`
reported "Wrote ..." whether or not anything did -- which looks like success
while the desktop stays exactly as it was.
It now checks the sibling cosmic.conf and prints the line to add when the file
is unreachable. The match is by filename and deliberately loose: it is looking
for evidence the user already knows about the file, not parsing the config.
The shipped template carries that `source` line commented out rather than
live, because `source` naming a file that does not exist is a hard error, and
a fresh checkout has no theme.conf yet. Copying the template and running
`apply` has to work before any theme is imported.
Three things stood between `assets.rs` and a themed desktop.
`import-theme` never called it. The module was written, tested and unreachable;
`--assets` now wires it up, with `--source`, `--overwrite` and `--dry-run`, and
finds the theme repo's Source/ directory by searching upward rather than
assuming HyDE's exact nesting depth.
The archive guard rejected every real icon theme. Refusing any `..` in a link
target is right for an entry path but wrong for a symlink: icon themes are
built out of relative links into sibling directories, and Tela ships thousands
of `../devices/network-wireless.svg`. What matters is whether the target
resolves inside the destination, which `stays_within_root` now decides
lexically -- no canonicalize, since the tree does not exist at plan time and
following real links during validation would be a TOCTOU window. Absolute
targets and links that climb past the root are still refused; the existing
escape tests still pass.
`apply` silently ignored `source`. It parsed and resolved inline while `watch`
went through `compile`, and `flatten` drops `Item::Source` -- so an include
that worked under `watch` vanished under `apply`. `apply` now uses `compile`
too. This matters immediately: the generated theme lives in its own
theme.conf, sourced from cosmic.conf, so re-importing a theme cannot clobber
the keybindings.
The waybar stylesheet claimed a theme could be dropped in ahead of it to
recolour the bar. It could not -- HyDE names its colours main-bg/wb-act-bg and
the rules referenced bar-bg/accent. Split into palette + theme + bridge +
rules, imported in that order, so the claim is now true. Verified by loading
the result through GTK's own CSS parser: with Tokyo Night installed main-bg
resolves to #24283b and wb-act-bg to #bb9af7; with an empty theme.css the
defaults stand. Both parse without error.
Two deliberate departures, both commented where they are made: the theme's
near-transparent bar-bg is composited at 0.85 because cosmic-comp has no blur
to put behind it, and theme.css is copied next to style.css rather than
imported from HyDE's own path, because a missing @import is fatal in GTK and
would break the bar on any machine without a theme.
The hyprcosmic profile does not start cosmic-bg, so until now nothing was
drawing a background at all.
HyDE calls this swww, and it is packaged for Fedora after all -- the
alebastr/sway-extras COPR carries it. Upstream renamed the project to awww at
0.12 and the package Obsoletes swww < 0.12.0, so `dnf install swww` lands
awww-0.12.1. /usr/bin/swww still exists as a shim, but it prints a deprecation
warning on every invocation and its own help says it will be removed in a
future update, so the autostart line uses the real name.
`bind = SUPER, D, exec, rofi -show drun` is the most recognisable line in a
hyprland.conf, and the hyprcosmic profile makes it necessary rather than just
idiomatic: with cosmic-launcher and cosmic-app-library not running, COSMIC's
stock Super, Super+/ and Super+A bindings point at nothing.
Binds are the one repeatable key in the language -- many lines fold into a
single map instead of the last one winning -- so they bypass the schema, which
is built around one conf key naming one value. They land in the Shortcuts
`custom` key, which cosmic-comp merges over `defaults`, so the system file is
untouched and reverting means deleting the lines and re-applying.
Actions are rendered as RON text rather than modelled as an enum: COSMIC's
Action has forty-odd variants, this crate deliberately does not link the cosmic
crates, and the mapping table only ever needs a handful. Dispatchers without a
genuine equivalent are refused rather than approximated, since a keybinding
that silently does the wrong thing is worse than one that fails to compile.
Verified the emitted file deserializes into cosmic-settings-config's own
`Shortcuts` type: five bindings, keysyms XK_a/XK_slash/XK_Return, Spawn actions.
The profile disables cosmic-panel, so without these a HyprCosmic session is a
compositor with no bar, no clock and no tray. Waybar's shipped default is not
a substitute: /etc/xdg/waybar/config.jsonc is built entirely from sway/*
modules and renders nothing under cosmic-comp.
Module choices are pinned to what cosmic-comp actually advertises on the
Wayland registry, checked against a live session rather than assumed:
ext/workspaces ext_workspace_manager_v1, present in stock cosmic-comp.
wlr/taskbar zwlr_foreign_toplevel_management_v1, absent from stock
cosmic-comp and supplied by this fork's Patch A. Under a
stock compositor the module stays empty and the bar
otherwise works, so the config is not fork-only.
tray cosmic-panel normally hosts the StatusNotifierWatcher via
cosmic-applet-status-area; with the panel disabled waybar
hosts it instead.
The stylesheet keeps every colour behind an @define-color so that a HyDE
theme's waybar.theme -- which is only a list of such declarations, the bespoke
CSS being HyDE's own rather than any theme's -- can recolour the bar by being
sourced ahead of it.
Configs install to /usr/share/hyprcosmic/waybar so the autostart file needs no
per-user paths, which matters because the profile parser is deliberately not a
shell and cannot expand ~ or $HOME.
Verified: config.jsonc parses as JSONC and every compositor module placed on
the bar has a matching config block.
Running a second cosmic-session on the development machine turns out to be
genuinely dangerous, and twice it logged the developer out mid-session and
destroyed open work. Two distinct causes, both encoded here as guards:
- Name-based process selection cannot distinguish the fork, the system
install, or a stand-in binary; they all answer to `cosmic-session`. The
second logout came from `pgrep -x cosmic-session | head -1` inside the
test written to demonstrate that name matching is unsafe, because `head
-1` favours the oldest match, which is always the live desktop. The
harness therefore never selects a process by name: it spawns under
setsid and signals `-$PGID`, with the group ID taken from `$!`.
- A nested cosmic-session takes the well-known D-Bus name
com.system76.CosmicSession away from the running session on a shared bus
("Connection `:1.3` lost name ..." in the journal), destabilising the
outer desktop before anything is killed. The harness always runs under
dbus-run-session. Nesting cosmic-comp alone does not need this.
It also refuses to start without WAYLAND_DISPLAY, since the winit backend
would otherwise fall back to DRM and seize the real display, and it reaps
IPC socket directories whose owning PID is gone.
Verified: shellcheck-clean syntax; the no-WAYLAND_DISPLAY guard fires; an
audit confirms every `kill` targets the script's own process group and no
code path matches a process by name. NOT verified: the harness has never
been run against the real binaries. Session-level runtime testing is now
deferred to a VM or to logging into hyprcosmic.desktop directly, rather than
nesting inside the developer's live desktop.
cosmic-session joins cosmic-comp in .gitignore; both are forks that become
submodules under the topology in the design spec.
watch: source-include expansion by textual splicing so diagnostic spans stay
correct across merged files; debounced inotify; a bad edit prints diagnostics
and keeps watching rather than killing the daemon.
assets: plan/apply split mirroring emit.rs. Tarball entries and symlink/
hardlink targets are validated before extraction, sharing one routine between
plan and apply so the check cannot drift.
Added tests/archive_escape.rs as independent verification of that boundary.
The tar crate refuses to build hostile archives through its safe API, so the
fixtures write GNU header name/linkname bytes directly — the same thing a
malicious archiver does. Asserts on the filesystem afterwards rather than on
returned errors, and covers symlink indirection, where neither entry path
contains '..' yet a later write still escapes.
99 tests.
import-theme translates a HyDE hypr.theme into cosmic.conf, reusing the
Phase 1 parser — which is the payoff for choosing Hyprland-style syntax.
Nothing is dropped silently. Every source key either lands in the output or
carries a Note explaining why not, classified as NoEquivalent,
NeedsCompositorPatch, DifferentProgram or Lossy. Gradient borders contribute
their first stop as the accent and say so.
Handles real-world quirks found in actual theme files: HyDE's |> destination
header (no '=', would otherwise be a parse error), colon-keys like
shadow:enabled, and nested blur blocks.
Tests run against the verbatim Catppuccin-Mocha theme, and assert that the
generated conf both parses and resolves against the registry. Verified
end-to-end on Tokyo-Night, a theme absent from the tests: import -> apply
produced 8 correct cosmic-config files with 20 settings reported.
64 tests.
emit writes RON directly rather than linking libcosmic. Spike 2 showed
cosmic-config is a filesystem KV store whose notify watcher keys off file
paths (lib.rs:377), so an atomic write is observed identically to the typed
API — for the cost of ron instead of the whole libcosmic graph.
Two-stage: plan renders without touching disk, apply writes. Composites are
read-modify-write against verified upstream defaults — gaps (0,8) at
theme.rs:939, CornerRadii at corner.rs:20-31 — so setting one field never
drops its siblings. Unmodelled composites error rather than write blind.
52 tests. CLI verified end-to-end: apply, --diff (writes nothing), idempotent
rerun, partial update preserving siblings, and multi-diagnostic failure with
exit 1 and zero writes.
Spike results corrected the spec: ThemeBuilder.gaps is (outer, inner) at
cosmic-theme/src/model/theme.rs:895, lives under CosmicTheme.{Dark,Light}.Builder
rather than CosmicTk, and fans out to two components. Entry therefore carries
targets: &[Target].
34 tests, including the folding property that keeps gaps_out from clobbering
gaps_in. Bare #rrggbb colours rejected: # begins a comment, as in Hyprland.
HyDE-style desktop on cosmic-comp: single-file Hyprland-idiom config
compiler, wlr-foreign-toplevel + Hyprland-compatible IPC patches, waybar
in place of cosmic-panel.
gitops
Jeremy Soller