Comment #10 on issue 3498 by stephane.ducasse: remove:ifAbsent: in linked
list leads to interruption/sync problems
http://code.google.com/p/pharo/issues/detail?id=3498
adrian: I'll try that...
But I still don't understand why the change in Pharo that makes
LinkedList>>remove:ifAbsent: non thread safe can cause the problem since
this code is executed by the timerEvent process, which runs at the highest
priority. This process should never be suspended during the execution of
remove:ifAbsent:. What do I miss?
Andreas: The problem isn't thread-safety, at least in the classical
definition. What happens is that if you're removing processes by using
LinkedList>>remove: you are subject to a race condition where the semaphore
gets signaled *while* you are removing the the process. Obviously,
hillarity ensues at this point, which is why I made primitive suspend do
the Right Thing (i.e., remove the process primitively). There are two
parameters which affect if you're likely to see the effect or not: One is
the number of suspension points (real sends) in the method. The more you
have, the more likely you're affected. The second one is whether the method
can tolerate having the process removed "underneith its feet". Both are far
worse in Pharo.
