Marc Weber wrote:
The next I wasn't sure about was speed: Is it worth trying to avoid some mallocs or not.
That's the main motivation for the special handling of context-local heaps. I want this code to scale to many-core execution, and malloc() will inevitably incur some thread synchronization overhead, while parceling out chunks of a persistent thread-local heap is almost free.
To me, this implies that if you do anything that is probably going to trigger some thread synchronization operations anyway (say, any IO, or calling a beast like the Ruby interpreter), then the performance argument stops being compelling. If it's easy to register a "free" handler, then the programmer productivity argument also becomes void.
_______________________________________________ Ur mailing list [email protected] http://www.impredicative.com/cgi-bin/mailman/listinfo/ur
