Comment 10 for bug 1038034

(Just seeing the above patch now. Will try it out now...)

The attached standalone testcase just uses three threads:
we try to terminate T2 around the time when T1 exists,
this way, T3 is never woken up.

tests$ while sbcl --script wait-on-semaphore.script.lisp ; do true ; done
::: Running (SEMAPHORE-NOTIFICATION WAIT-ON-SEMAPHORE)
.-.
::: Success (SEMAPHORE-NOTIFICATION WAIT-ON-SEMAPHORE)
::: Running (SEMAPHORE-NOTIFICATION WAIT-ON-SEMAPHORE)
.-.
[ 0.1s] 1 safe remaining: T3
[ 0.2s] 0 safe remaining
::: Success (SEMAPHORE-NOTIFICATION WAIT-ON-SEMAPHORE)
::: Running (SEMAPHORE-NOTIFICATION WAIT-ON-SEMAPHORE)
.-.
::: Success (SEMAPHORE-NOTIFICATION WAIT-ON-SEMAPHORE)
::: Running (SEMAPHORE-NOTIFICATION WAIT-ON-SEMAPHORE)
.-.
::: Success (SEMAPHORE-NOTIFICATION WAIT-ON-SEMAPHORE)
::: Running (SEMAPHORE-NOTIFICATION WAIT-ON-SEMAPHORE)
.-.
[ 0.1s] 1 safe remaining: T3
*** Notes:
("T1" 0 71)
*** Inc's:
("T1" 0 81)
*** Kills:
("T2" T 81 81)
Backtrace for: #<THREAD "T3" RUNNING {1003694443}>
0: ("bogus stack frame")
1: ((FLET #:WITHOUT-INTERRUPTS-BODY-575 :IN CONDITION-WAIT))
2: (CONDITION-WAIT #<WAITQUEUE {1003693953}> #<MUTEX (free) {1003693923}> :TIMEOUT NIL)
3: ((FLET SB-THREAD::WITH-SYSTEM-MUTEX-THUNK :IN WAIT-ON-SEMAPHORE))