News for hackers. Stuff that matters.
Re:Ferret: An Experimental Clojure Compiler by oconnore on Thursday June 30, 2011 @09:46AM (#2713972)
I applaud his effort (and this certainly isn't an easy project, he's a smart guy, etc.), but felt progressively more and more queasy reading this. I thought I could keep going even once I had realized the necessary lack of a runtime lambda or first class closure. When I got to the refcounting part, I mashed the back button in horror. Am I spoiled or what?
Re:Ferret: An Experimental Clojure Compiler by nakkaya on Thursday June 30, 2011 @10:03AM (#2714038)
The reason I used ref counting instead of a proper garbage collector is that, most microconstollers have couple kilobytes of ram (a ATmega328 has 2KB) every garbage collection scheme I've looked at was using a lot of ram, plus ref counting is slow but it is predictively slow, if you are taking measurements at 100 times a second it will always be 100 times a second, you won't have pauses right in the middle of the routine (or say when you are in the air making a maneuver).
Re:Ferret: An Experimental Clojure Compiler by radarsat1 on Thursday June 30, 2011 @12:31PM (#2714687)
It's a myth that malloc/free is constant-time in a reference-counting scheme. If you are doing real-time programming, malloc/free should be avoided as much as garbage collection.