[phischdev]

checkIntDeps probes yad for its version and compares it against 7.2:

YADVER="$("$YAD" --version | tail -n1 | cut -d ' ' -f1)"

When the probe produces nothing, YADVER is the empty string — and sort -V puts that before 7.2, so a yad that simply would not answer is reported as

ERROR - checkIntDeps - Yad version '' is too old. You need to update to at least '7.2'

and the whole run exits.

This happens in practice

A systemd user unit started at login inherits a DISPLAY that does not answer yet. yad exits without printing, and the probe comes back empty. From the journal of a Steam Deck running the artwork watcher from #15:

2026-10-10T22:23:52  -- boot --
2026-10-10T22:24:23  Starting tinkergame-artwork.service...
2026-10-10T22:24:28  yad[5033]: cannot open display:
2026-10-10T22:24:28  ERROR - checkIntDeps - Yad version '' is too old. You need to update to at least '7.2'
2026-10-10T22:25:12  Starting tinkergame-artwork.service...
2026-10-10T22:25:15  ERROR - checkIntDeps - Yad version '' is too old.
2026-10-11T00:00:00  Starting tinkergame-artwork.service...   → fine

31 and 80 seconds after boot it fails; from then on it works. Ordering the unit After=graphical-session.target (which #20 did) is not enough, because the target goes active before a new client can open the display.

The command being aborted here needs no GUI at all — fetching artwork never opens a window. Any non-interactive use of TinkerGame from a unit, a hook or a script hits the same wall.

The change

An unreadable version is reported as unknown and the check is skipped for that run, instead of being read as "too old". Three cases now behave differently from each other:

Probe result	Before	After
empty (could not ask)	"too old", exit	warn, continue
read, below minimum	"too old", exit	unchanged
read, at or above minimum	proceed	unchanged

Only a version that could actually be verified is written back to the global config, so an unknown one does not get recorded as good.

Testing

Six tests in tests/unit/yadversion.bats, driving checkIntDeps with a yad stub that reports nothing, 6.0, or 9.3. Two fail on main and pass here; the other four pin the behaviour that was already correct, including that a genuinely old yad still aborts.

Note on CI

The Unit tests job fails on setGameVortexStaging: a configured stage is not replaced (tests/unit/vortex.bats:347). That is not from this change — it reproduces on an unmodified main in the same CI, while passing locally for me on both branches. I can open a separate issue if that is useful.