Home > java, scala, software engineering, Uncategorized > Java Concurrency Pitfalls (in Scala) Answers

Java Concurrency Pitfalls (in Scala) Answers

Recently, Cay Horstmann posted a dozen Java concurrency pitfalls ported to Scala for extra pitfalliness. I had fun figuring them out. Here’s what I came up with.

I won’t copy all the code here, just provide my answers.

  1. The var stop isn’t synchronized in any way (synchronized, AtomicBoolean, etc) so there’s a chance that when it’s set to false, the change won’t be seen in the thread.
  2. Here, stop is an AtomicBoolean, fixing the issue in problem #1. That’s good. However, if doSomething() throws an exception done.put("DONE") will never execute and done.take() will wait forever.
  3. Here, Thread.run() is called rather than Thread.start(). The so-called “BackgroundTask” will actually run in the “foreground”
  4. ConcurrentHashMap.keySet().toArray will give a “weakly consistent” snapshot of the keyset at the time of the call. By the time it returns, the keys in the map may have changed completely.
  5. Oops. A string literal is used for the lock meaning that all instances of Stack will most likely share the same lock.
  6. Oops. A new “lock” is created every time push() is called, which is just as good as no lock.
  7. Here values is mercifully synchronized, but the var size isn’t. That is, if size is used in any other methods it may not be synchronized correctly.
  8. A do/while is used for the condition variable rather than just while. If cond is in a signaled state initially, the condition that size is zero may not be checked.
  9. If out.close() throws an exception, the myLock.unlock() will not be executed.
  10. Here, a string is being added to a blocking queue within a UI event handler. If the queue is full, queue.put() will block causing the UI to become unresponsive.
  11. I’m not totally sure with this one. I can think of a few things that could go wrong. First, if a listener is added or removed while fireListeners() is executing, you could get a ConcurrentModificationException. I think so anyway. I wasn’t able to find any indication of how ArrayBuffer iteration handles this. Second, since the listeners are notified with the lock held, you’ll almost certainly get a deadlock eventually, usually a lock inversion with some other thread that’s calling SwingUtilities.invokeAndWait().
  12. Ridiculously, SimpleDateFormat is NOT THREADSAFE! So using the same formatter from multiple threads is a recipe for sadness.

I wonder what I missed.

Cheers!

java, scala, software engineering, Uncategorized , ,

  1. May 11th, 2011 at 01:51 | #1

    11 should use CopyOnWriteArrayList in order to allow a listener to call add/remove listener when the event is being dispatched without causing a CME and/or deadlock.

  2. Esko Luontola
    May 19th, 2011 at 13:04 | #2

    7 is synchronizing on a non-final variable. Not all method calls will be synchronized on the same lock.

  3. May 19th, 2011 at 13:27 | #3

    Esko. Good catch! Thanks.

  1. No trackbacks yet.