[concurrency-interest] ForkJoinTask enhancement suggestion

Doug Lea dl at cs.oswego.edu
Mon Mar 5 06:39:13 EST 2012


On 03/04/12 23:26, Wolfgang Baltes wrote:
>  My
> proposal is to consider adding a new constructor ForkJoinTask(V) that a) sets
> the value, and b) sets the state to NORMAL. Although it may be consistent to
> also add a constructor to create an "exception task", I do not think this is
> necessary because of less frequent use, and hence less overall performance impact.
>
> Noticing that tagging will also be available in a forthcoming version of the
> framework, it may make sense - for better performance and consistency - to also
> add the constructor ForkJoinTask(V, short) and the method complete(V, short),
> which allow setting both the value and the tag simultaneously.
>

Stay tuned. One of the reasons for adding integrated tagging support
is to allow introduction of one or more new flavors of
ForkJoinTask (besides RecursiveAction and RecursiveTask)
that trigger completion actions when some number (including zero)
of subtasks complete. So you should be able to efficiently get these
kinds of effects using these classes without us having to expose
more representation-dependent methods/constructors.

These upcoming classes are part of the efforts I mentioned in posts
around 6 months ago to improve support for the many FJ usages
that have sprung up that fall outside of the initial target
uses.

-Doug




More information about the Concurrency-interest mailing list