I'm not exactly sure on the reproduction scenario here, but I have had
switching ruleset with converts disabled crash on me a few times
today. It appears to happen sometimes when after the switch the expanded
group no longer exists in the set mapping, because a filter just ran and
removed that group from set of possible groups because there'd be no
beatmaps under it.
I tried to manufacture a quick test but it's not a quick one to test
because filtering intereference is required to reproduce, I think.
I will try again on request but I mostly just want to get this fix out
ASAP before I finish up for the day.
Because the detached store exists and has a chance to actually
semi-reliably intercept a beatmap update operation, I decided to try
this. It still uses a bit of a heuristic in that it checks for
transactions that delete and insert one beatmap each, but probably the
best effort thus far?
Notably old song select that was already doing the same thing locally to
itself got a bit broken by this, but with some tweaking that *looks* to
be more or less harmless I managed to get it unbroken. I'm not too
concerned about old song select, mind, mostly just want to keep it
*vaguely* working if I can help it.