[GNOME 50 / EL10] org.gnome.Shell.target missing — gnome-session-init-worker aborts on first boot

[hanthor]

Background

While triaging tuna-os/github-copr#15, we identified that gnome-session-init-worker aborts on first boot because it cannot find org.gnome.Shell.target. An attempt was made to fix this by uncommenting the unit in gnome-shell.spec, but investigation showed:

• org.gnome.Shell.target is not present in the GNOME 50.0 tarball — meson does not install it
• The line is also commented out in the Fedora 44 spec (#{_userunitdir}/org.gnome.Shell.target)
• The unit was present in EL10 AppStream's GNOME 47/49 builds

The revert is tracked in tuna-os/github-copr@469fd1c.

The Problem

gnome-session-init-worker gnome-initial-setup crashes with SIGABRT because it expects org.gnome.Shell.target to exist as a systemd user unit. Without it, the GDM greeter session cannot register and the system hangs at a blank screen on first boot.

What We Don't Know

Where is this unit supposed to come from in GNOME 50? Possible explanations:

1. Removed upstream intentionally — GNOME 50 may have restructured session startup so this target is no longer needed, and gnome-session-init-worker should not be looking for it. Could be a gnome-session bug that needs an upstream fix or backport.

2. Moved to another package — It may have migrated to gnome-session or a new session-related package that we aren't shipping yet.

3. Needs to be a static file in gnome-shell — The .target file may not be generated by meson but should be shipped as a static source file. Worth checking whether the Fedora 44 gnome-shell package has it in the installed RPM even though meson doesn't build it (i.e., installed via a Source: + manual install in the spec).

4. gnome-session-init-worker has a graceful fallback on Fedora — Fedora 44 may ship a patched gnome-session that tolerates the missing target, and EL10 does not.

Suggested Investigation Steps

• rpm -ql gnome-shell on a working Fedora 44 system — does org.gnome.Shell.target appear?
• If yes: check if it comes from a Source: file in the Fedora spec or is installed some other way
• Check gnome-session upstream git log around the GNOME 50 timeframe for changes to gnome-session-init-worker and how it handles missing units
• Search gnome-session source for org.gnome.Shell.target — is the worker hardcoded to require it?
• Check if gnome-session-init-worker logs a specific error vs. unconditional abort — a graceful fallback patch may exist
• Check EL10's stock gnome-session (GNOME 47) — does it ship a org.gnome.Shell.target that our GNOME 50 session is accidentally inheriting/expecting?

Related

• [GNOME 50] gnome-session-init-worker crashes due to missing units and SELinux denials github-copr#15 — original crash report
• If gnome-initial-setup is present, the system just hangs on first boot github-copr#2 — first-boot hang (same root cause)
• Commits fixing the SELinux and gnome-initial-setup parts of Add just lint to git precommit #15: d9b8ee5

[hanthor]

Root cause found via rpm -ql gnome-shell on ghcr.io/ublue-os/bluefin:beta (gnome-shell 50.0-1.fc44):

org.gnome.Shell.target does not exist on a working Fedora 44 / GNOME 50 system. It is not shipped by gnome-shell, gnome-session, or anything else. No session file references it either.

The crash is a version mismatch: EL10's stock gnome-session 46.0 ships a gnome-session-init-worker that still references this target (removed in GNOME 50). Our gnome-session 50.0 COPR package was missing Obsoletes: gnome-session < 50.0, so on systems where EL10's gnome-session was already installed, the old binary could remain on disk alongside our GNOME 50 packages.

Fix: 959edd5 adds Obsoletes: gnome-session < 50.0 to our spec, ensuring RPM replaces the EL10 46.0 package on upgrade. gnome-session build queued in both COPRs.