On Tuesday, October 9, 2012 9:37:20 AM UTC-4, tbc++ wrote: > > >> Polyfns are exactly as fast as protocol functions (I am using the same > >> caching and dispatch code), but they do not generate a Java interface, > >> and they are slightly simpler to define. > > So I guess I have to ask the question again...what is the true use case of > polyfns? Are they faster than multimethods? >
Dramatically faster than multimethods, just like non-interface based protocol dispatch. It is the same code. > If I can get a dramatic speed up by using interfaces, why would I throw > that away? > Sorry, I should have said that polyfns are exactly as fast as non-interface protocol dispatch. As with everything, it's all about tradeoffs. I don't recall that I said everyone should be replacing their uses of protocols with polyfns, so I don't know why I'm supposed to convince you to do so. I think it is useful to have fast polymorphic dispatch separate from class generation, and useful as an orthogonal building block with granularity at the function level. If you're interested in pure performance, then feel free to use protocols. There are downsides to that approach, too. I think maybe polyfns might be fast enough for most use cases, but I have some ideas for improving performance, too. I've been experimenting with some Java 7 features. I have a stashed version of polyfns that use java.lang.ClassValue as a cache for implementation fns instead of a clojure.lang.MethodImplCache. The performance is pretty good, and the implementation is simpler. I'm wondering what kind of performance I could eek out by having a ClassValue cache that caches MethodHandles to the implementations instead of the IFns themselves. Obviously Clojure isn't emitting invokedynamic call sites, but I wonder whether the JVM could inline through that kind of a call chain. Cheers, Paul -- You received this message because you are subscribed to the Google Groups "Clojure" group. To post to this group, send email to clojure@googlegroups.com Note that posts from new members are moderated - please be patient with your first post. To unsubscribe from this group, send email to clojure+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/clojure?hl=en
