I've made my own take of your plural dispatch system. https://gist.github.com/3659663
I switched from using vars and a map for determining which pluralfn to use, to atoms and metadata tied to the pluralfn. My method reduces the amount of indirection and invocations. I originally tried to make a deftype version to store the data, but then I'd have to reimplement all of the microsyntax of of clojure.core/fn and expand out the args to all the arities, not fun. Also I created defplural-body which doesn't specify the argument arity which you can use to avoid sequencing/unsequencing arguments for performance. I defined defplural in terms of defplural-body, which maintains compatibility with your examples. My code has the advantage over some implementation details in multimethods in that redefining the pluralfn does not cause you to forget all the implementations since defplural-body checks for existing implementations. This, however, will require the user to run clear-implementations! if one does want to forget all the implementations. On Wednesday, September 5, 2012 6:48:36 AM UTC-4, Chris Ford wrote: > > I was watching David Nolan's talk about predicate > dispatch<http://blip.tv/clojure/david-nolen-predicate-dispatch-5953889>, > and he articulated something that I've found frustrating about multimethods > - the set of implementations is open, but the dispatch function is closed. > That means that if your dispatch value can't be computed in the same way > for all your data, you can't use multimethods. > > Predicate dispatch allows implementations to be registered with a test > that decides whether or not they are applicable, so it can do things that > multimethods cannot. For example, predicate dispatch would allow you to > write an open function that categorises integers as "prime", "perfect" or > some other category without baking them all into the dispatch function. > > In his talk, David speaks about the problem of overlap. 2 is both "prime" > and "even", for example. That got me thinking whether overlap is a bug or a > feature. > > I've hacked together a simple plural dispatch > system<https://gist.github.com/3634871>, > which allows the specification of a custom resolution function that can > decide whether to use one or all of the registered implementations. You can > use this system to implement both multimethods and predicate dispatch. > > What practical reason would you have for doing this? Random dispatch! > Can't decide which implementation to use? Now you don't have to! > > (defplural random (fn [implementations args] (-> implementations rand-nth > (apply args)))) > (defimplementation random #(+ % 100)) > (defimplementation random #(- % 100)) > (random 1) > > It could also be useful for gathering a set of all possible results, > quasi-AOP extension points, broadcasting events etc. > > Cheers, > > Chris > -- 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
