[18:08:38] <smcv> grundleborg: oh I see what the problem is
[18:09:03] <smcv> grundleborg: it decides what the possible handlers are up-front, at the time the request starts
[18:09:20] <smcv> grundleborg: on the assumption that the final channel will match the request
[18:09:32] <smcv> grundleborg: Requested=true isn't in the request :-(
[18:09:35] <smcv> grundleborg: (and can't be)
[18:10:04] <grundleborg> eek :'(
[18:10:09] <smcv> grundleborg: this badly needs fixing, it's a special case of "should run preferred handler even if we don't have any idea what it is or does"
We should try to get this merged for 5.3.0.
Fixed in 5.3.0, might be worth backporting to 5.2