fix: session edge paths, non-well-formed locked rc.xml, fail-soft packaging#4
Open
DeBondor wants to merge 5 commits into
Open
fix: session edge paths, non-well-formed locked rc.xml, fail-soft packaging#4DeBondor wants to merge 5 commits into
DeBondor wants to merge 5 commits into
Conversation
The header comment mentioned singularity-shortcut's --serve mode. "--" is not allowed inside an XML comment, so libxml2 fails the whole file with "Double hyphen within comment" and labwc discards the locked config: the greeter and OOBE compositor run with labwc defaults, without the keybind lock this file exists to apply.
In client mode the OOBE runs as root and pointed XDG_RUNTIME_DIR at /run/singularity purely to reach the persistent compositor's socket. Every client library root loads then puts its runtime state there: at-spi, dconf, gvfs and gvfsd land as root-owned 0700 directories, and greeter-ready as root:root 0644, in a directory owned by the greeter (0770 greeter:singularity-session) that the greeter has to write once we hand over. The greeter runs as uid greeter and cannot rewrite any of it, so it dies on "can't create /run/singularity/greeter-ready: Permission denied" and greetd loops on "greeter exited without creating a session": a black screen after the OOBE completes. Keep XDG_RUNTIME_DIR at root's own /run/user/0 and reach the compositor through an absolute WAYLAND_DISPLAY, which libwayland uses verbatim instead of resolving against XDG_RUNTIME_DIR. greeter-ready is the one file that still has to be shared, so mark it with chmod as the own-labwc fallback already does. The fallback starts its own labwc, which binds its socket itself, so the inherited absolute path is unset there.
The installer's last page offers a button labelled "Restart now", but its
handler is only `this.close ()`. atom-install-session ran labwc with -s, so
the compositor outlived the window it was started for: the user pressed
"Restart now" and got a black screen with a cursor, with nothing left to
interact with and no way out short of a hard power cycle.
Make the exit path explicit in the launcher, which is the only place that
knows what should happen next:
- atom-install drops /run/atom/install-complete. singularity-installer
exits 0 both when the install finished and when the window was closed
early, so exit status alone cannot tell the two apart.
- atom-installer-launch stops exec'ing the installer and acts on its
exit: reboot on a finished install (what the button promises), return
on a finished OOBE (greetd moves on to the greeter), otherwise
relaunch the wizard up to three times.
- both session scripts run labwc with -S instead of -s, so the
compositor terminates with the launcher and no clientless compositor
is ever left on screen.
Verified on a live boot: the "is ready" page stays up until the button is
pressed, then the machine shuts down cleanly (stopping default.target,
sync, rebooting) and comes back into the OOBE on the installed disk.
The cmd/fw-verify guard covers an AtomLoops checkout that has no such directory, but not one where the directory exists and the build fails (a mid-refactor HEAD, a dependency it cannot resolve offline). Under set -e the bare go build then aborts the whole run, after the erofs, the verity tree and the UKI are already done. build-initramfs.sh already treats a missing verifier as "no firmware verifier, base firmware only", so this is a degraded image rather than a broken build. Warn and carry on.
Buildroot 2026.02.2, which prepare.sh downloads, ships
gsettings-desktop-schemas 45.0, whose org.gnome.desktop.interface has no
accent-color key. That key landed in 47.
singularity-shell writes it unconditionally on startup. main.vala:100
registers an Idle callback that calls update_accent_color(), which reads
dev.sinty.desktop accent-color -- default 'blue', a named colour, so the
"skip wallpaper-derived and custom colours" guard does not apply -- and
then does, at main.vala:883:
try {
var iface = new GLib.Settings("org.gnome.desktop.interface");
iface.set_string("accent-color", color_name);
} catch (GLib.Error e) {
The catch cannot help. A key absent from the schema is a g_error inside
GIO, not a returned GError, so the process dies rather than raising:
$ gsettings list-keys org.gnome.desktop.interface | grep -c accent-color
0
$ python3 -c 'from gi.repository import Gio
s = Gio.Settings.new("org.gnome.desktop.interface")
try: s.set_string("accent-color", "blue")
except Exception as e: print("caught", e)'
GLib-GIO-ERROR **: Settings schema 'org.gnome.desktop.interface' does
not contain a key named 'accent-color'
$ echo $?
134
134 is SIGABRT. singularity-desktop dies on every shell start, the
session supervisor gives up and greetd falls back to the greeter, which
is a login loop.
Written as buildroot-source-patches/0006-*.patch to match the patch set
eec8a89 introduced. Hashes are the upstream sha256sum file for 48.0.
Verified that it applies forward on a pristine 2026.02.2, that a reverse
dry-run is rejected there, and that the reverse dry-run succeeds on an
already-patched tree so prepare.sh reports "already applied" on a rerun.
DeBondor
force-pushed
the
fix/session-and-packaging-edges
branch
from
July 26, 2026 11:18
3eb450f to
db373a1
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Five independent fixes on top of
eec8a89, all on edge paths that the happypath never reaches. None of them changes install-and-first-boot behaviour.
I booted the
nightlyISO in QEMU with OVMF and swtpm first, so the overlap withyour own path-and-signing rework is already resolved: the
/home/mirko/...defaults and the sibling-checkout resolution are yours, not duplicated here.
Install → firstboot → greeter → desktop all worked, and the only error on the
serial console was
bluetooth.service: exited before acquiring bus name org.bluezon the live boot, which nothing here touches.fix: drop the double hyphen from the labwc-locked rc.xml commentlabwc-locked/rc.xmlis not well-formed XML onmain. The comment block thatopens on line 2 runs to line 25, and line 21 mentions
--serveinside it, whichXML forbids:
The greeter came up fine in the nightly, so libxml2 is evidently recovering from
it, but that is not something to rely on. Reworded the prose so the file parses.
fix: keep the OOBE's runtime state out of the greeter's XDG_RUNTIME_DIRatom-oobe-sessionexportsXDG_RUNTIME_DIR=/run/singularityfor its owncompositor, which is the greeter's directory. Its runtime state then lands in a
directory owned by another user and, per the comment already in
singularity-session-common, a failed log open there is exactly the case thatleaks onto the seamless boot.
fix: reboot after the installer instead of leaving a bare compositorsscommon_exec_labwc -sleaves labwc running after the startup command exits, soafter the installer's final "Restart now" button (which only closes the window)
the user is left looking at an empty compositor.
-Sterminates it with thecommand.
atom-installer-launchreboots on a finished install, so this onlymatters on the give-up path.
Merged cleanly with the splash / DRM-master handoff you added to
atom-install-sessionineec8a89.build: do not abort packaging when fw-verify fails to build[ -d "${ATOMLOOPS}/cmd/fw-verify" ]covers a checkout without that directory,but not one where the directory exists and the build fails. Under
set -ethebare
go buildthen aborts the whole run, after the erofs, the verity tree andthe UKI are already done.
build-initramfs.shalready treats a missing verifieras "base firmware only", so warn and carry on instead.
build: bump gsettings-desktop-schemas to 48.0 for accent-colorThis one needs the most justification, because the nightly does not exhibit
it — see the last paragraph for why.
prepare.shdownloads Buildroot 2026.02.2, which shipsgsettings-desktop-schemas 45.0. Confirmed straight from the release tarball your
own script fetches:
accent-colorlanded in schemas 47.singularity-shellat the pinnededab6504writes it unconditionally on startup:src/core/main.vala:100registers an
Idlecallback that callsupdate_accent_color(), which readsdev.sinty.desktop accent-color. That defaults to'blue'insingularity-desktopat the pinnedc7a6f3d4, a named colour, so the"wallpaper-derived and custom colours have no named equivalent; skip syncing"
guard does not apply, and
main.vala:883runs:The
catchcannot help: a key absent from the schema is ag_errorinside GIO,not a returned
GError, so the process dies rather than raising. Reproducedagainst a schema carrying the same id and no
accent-colorkey, with the sametry/catch shape:
134 is SIGABRT, and nothing was caught. So on a clean checkout
singularity-desktopdies on every shell start, the session supervisor gives upand greetd falls back to the greeter: a login loop.
Why the nightly does not show this. Its image already ships schemas 48.0. I
carved the erofs out of
sinty-os.isoand diffed the schema:It is byte-identical to the 48.0 tarball, and the image also carries
org.gnome.desktop.break-remindersandorg.gnome.desktop.screen-time-limits,both of which are absent from 45.0. Since nothing on
mainbumps the package,that image was built against a
buildroot-srcthat already carried the bumprather than one
prepare.shproduced from scratch — worth a look on its own,because it means the nightly is not currently reproducible from a clean
checkout of
eec8a89.Written as
buildroot-source-patches/0006-*.patchto match the patch seteec8a89introduced. Hashes are the upstreamsha256sumfile for 48.0.Verified that it applies forward on a pristine 2026.02.2, that a reverse
dry-run is rejected there, and that the reverse dry-run succeeds on an
already-patched tree, so
prepare.shreports "already applied" on a reruninstead of failing.