`setup/service.ts` decides between a real systemd user unit and the
nohup fallback by running `systemctl --user daemon-reload` and watching
for the call to succeed. In contexts that bypass pam_systemd —
`su -`, `pct enter`, headless containers — `XDG_RUNTIME_DIR` and
`DBUS_SESSION_BUS_ADDRESS` are not exported, the probe fails with
`Failed to connect to bus: No medium found`, and we install nohup
despite the user manager being healthy on disk.
Before the probe, check whether linger is enabled for the current user
and `/run/user/<uid>` exists; if so, re-derive the env vars from disk
and let them propagate to subsequent `systemctl --user` calls via
`process.env` (execSync inherits by default). If the probe still fails
after that, the existing nohup fallback runs unchanged — and the
warning log now records *which* precondition failed so the cause is
visible without grepping setup.log.
The pure decision function lives in `setup/systemd-user-env.ts` so it
can be tested without execSync. New regression test in
`setup/systemd-user-env.test.ts` covers the #2482 repro plus the
already_set / no_linger / no_runtime_dir / no_user / no_uid branches.
Closes#1981Closes#2482
glifocat