Comments on: Bringing ‘yield return’ from C# to Java

By: jimblackler

jimblackler — Fri, 01 Jun 2012 12:10:11 +0000

Thanks Olivier, consider it public domain.

By: Olivier Mélois

Olivier Mélois — Fri, 01 Jun 2012 09:51:39 +0000

This is absolutely interesting.

Have your code been published under an open-source license ? I’m working on a project to allow the creation of easily customizable model iterators. The project has not really started yet, but it will in a few days. I’m thinking about re-using some of your code, would you allow it.

Great article :D

By: Chris

Chris — Tue, 12 Apr 2011 17:56:09 +0000

I don’t know if you’re interested, but I made a quick change to allow runtime exceptions thrown in a collect method to be re-thrown by the threaded yield adapter.

diff -r 0107610716ee src/net/jimblackler/Utils/ThreadedYieldAdapter.java
— a/src/net/jimblackler/Utils/ThreadedYieldAdapter.java Tue Apr 12 13:42:43 2011 -0400
+++ b/src/net/jimblackler/Utils/ThreadedYieldAdapter.java Tue Apr 12 13:51:27 2011 -0400
@@ -27,6 +27,16 @@
private class AbortedMessage extends StopMessage {

}
+
+ private class ExceptionMessage extends StopMessage {
+ RuntimeException cause;
+ ExceptionMessage(RuntimeException cause) {
+ this.cause = cause;
+ }
+ RuntimeException getCause() {
+ return cause;
+ }
+ }

/**
* The vehicle to pass the actual values.
@@ -103,6 +113,8 @@
// to receive it, and the thread will block.
synchronousQueue.put(new AbortedMessage());
}
+ } catch (RuntimeException e) {
+ synchronousQueue.put(new ExceptionMessage(e));
}

} catch (InterruptedException e) {
@@ -140,6 +152,9 @@
try {
returnQueue.put(new Object()); // allow other thread to gather result
messageWaiting = synchronousQueue.take();
+ if (ExceptionMessage.class.isAssignableFrom(messageWaiting.getClass())) {
+ throw ((ExceptionMessage)messageWaiting).getCause();
+ }

} catch (InterruptedException e) {
messageWaiting = new EndMessage();

By: Neal Gafter

Neal Gafter — Wed, 05 Nov 2008 05:30:01 +0000

See http://gafter.blogspot.com/2007/07/internal-versus-external-iterators.html for another take at this problem.

By: Howard Lovatt

Howard Lovatt — Wed, 15 Oct 2008 09:23:49 +0000

I think that Phasers (http://www.cs.rice.edu/~vs3/PDF/SPSS08-phasers.pdf) are going to be part of Java 7 and I think these can achieve the same aim of generating a list lazily, but via a slightly different paradigm.

By: jimblackler

jimblackler — Sun, 12 Oct 2008 18:56:44 +0000

Thanks a lot for your comments Aviad. And also for your original article which I found very interesting. I confess I haven’t profiled my implementation for performance, and it does seem likely this may make this technique unsuitable for some applications.

By: Aviad Ben Dov

Aviad Ben Dov — Sun, 12 Oct 2008 18:12:27 +0000

This is great. My first idea was to use a threading system like yours, and I eventually reasoned to create the yielder the way I did for two reasons:
1. It was more efficient, resource-wise. When dealing with tree-searching model, for example (see my collections example), you can find yourself in a canonical-yielder situation. With a threading model, you could quickly use the computer’s resources. You end up with a lot of locks (due to the SynchronousQueue) and a lot of events running around your application.
2. It opens the path to adding a mechanism to a pre-process compiler (like APT), or even to OpenJDK, to be done in the C# way: in compile-time. The threading method works but its a workaround for a missing feature.

Don’t get me wrong: effectively, the threading solution works (as mine is broken) and it probably takes less time to develop and understand, so kudos! :)