On Sun, Aug 21, 2011 at 11:27 PM, Brent Millare <brent.mill...@gmail.com>wrote:
> I have a question about the presentation. > > You mention that you can't achieve the same dispatching performance in > the open case compared to the closed space. > I meant to say that the problem is hard. No completely unsolvable. > In my imaginary implementation, extend-pred takes the predicate label, > the pattern, and the code to be executed (note its code, not a > function, so you lose the ability to make closures). Each time extend- > pred is run, a new dispatch tree is compiled with the code to be > executed integrated into the tree. (This would be done by saving (and > updating) the code and patterns in a data structure that gets parsed > and generates code that is the match macro wrapped into a function.) > It's certainly a possibility. I haven't ruled it out yet. I'm mostly concerned with explosive code generation. match by taking the decision tree approach already favors performance over code size. There are patterns that are exponential in the amount of code they generate. Fortunately Maranget did some measurements with real world patterns - code size for decision trees built with good heuristics were never more than 1.5-2X the size of their backtracking counterparts. For pattern matching code size is a one time cost. For predicate dispatch, that's a lot of code to generated, since every new predicate case will produce an entirely new tree. But perhaps people won't care that much. Only time and experience reports will tell. David -- 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
