14:56:28stassatserror is known to return NIL at ir1 conversion, but at that time there's no subsequent blocks to unlink
14:58:58stassatsand multiple-value-bind obscures things enough
15:06:04stassats(lambda () (logtest ((lambda (v &rest args) (setf v ((lambda (&rest args) (error "")) v))) 1) 1))
15:11:38stassatsactually, when it propagates the SET is already dead
15:12:43stassatsalright, a check for node-to-be-deleted-p should do, for now
15:17:03stassatsit's still a bit weird, how come the SET gets derived as NIL, why isn't it discarded?
15:17:18stassatsoh no, i should stop while i'm ahead
15:28:12stassatsok, that one is clear, but now, how does a deleted block is followed by a live one?
15:30:58stassatsanother way to fix it is to remove (block-delete-p block) in maybe-terminate-block
15:46:42pfdietzThis showed up now because I added error forms to the random form generator. They throw a recognizable error that is considered not a bug if it happens in both evaluations.