On Mon, 2014-02-03 at 00:00 +0000, Cooler wrote: > I have several tasks. Each task may or may not create another > task. What is the best way to wait until all tasks finished?
What you are describing here is a classic fork/join architecture. The tasks are structured as a tree with synchronization handled by the sub-nodes. As far as I am aware std.parallelism focuses on data parallelism which is a scatter/gather (aka map/reduce) model of just a single layer. All the code fragments in the thread have, I believe, been predicated on working with a thread pool as an explicit global entity. I think the problems have stemmed from taking this viewpoint. I would suggest following the way the Java fork/join framework (based on Doug Lea's original) works. There is an underlying global thread pool, but the user code uses the fork/join abstraction layer in order to create the tree of synchronization dependencies. In this case instead of working with tasks directly there needs to be a type whose job it is to be a non-leaf node in the tree that handles synchronization whilst nonetheless creating tasks and submitting them to the pool. This is clearly something that could turn into an addition to std.parallelism or be std.forkjoin. Sorry I have no actual code to offer, but the overall design of what is needed is well understood, at least in the Java context. C++ has a long way to go to catch up, as does D. The other thing that then sits on this is lazy stream parallelism, which is what Java 8 is adding to the mix. -- Russel. ============================================================================= Dr Russel Winder t: +44 20 7585 2200 voip: sip:russel.win...@ekiga.net 41 Buckmaster Road m: +44 7770 465 077 xmpp: rus...@winder.org.uk London SW11 1EN, UK w: www.russel.org.uk skype: russel_winder
