moribund threads initiating GC
Moribund threads have been reported initiating GC.
Part of the problem is easy to fix: bind *current-thread* in the C-side
trampoline so that it remains valid after the lisp code for the thread
has finished (and check it in sub-gc.)
Understanding why and how a moribund thread can enter SUB-GC
would be crucial, however.
Sadly, I have not been able to replicate this for a while. This is
sufficiently bogus, however, that diagnosing the problem in depth
on a legacy SBCL is probably worthwhile.