2011/4/10 Eli Barzilay <e...@barzilay.org>: > 25 minutes ago, Jens Axel Søgaard wrote: >> 2011/4/10 Shriram Krishnamurthi <s...@cs.brown.edu>:
>> >> One advantage with the LLVM solution is that one is sure that the >> >> semantics of the parts of Racket that are implemented in C will >> >> be preserved. I am thinking such things as the numerical tower, >> >> whose C implementation contains quite a few functions that are >> >> non-trivial to implement directly in JavaScript. >> > >> > I don't know what the porting effort is to get Racket to LLVM. >> >> I hope that's not neccessary. Here is what I had in mind: GCC has an >> LLVM backend. > > (Last time I checked it was the other way around: LLVM had a GCC > front-end.) You are right. >> So compile Racket (with a disabled JIT) to racket.llvm with GCC, and >> then compile it with Emscripten. > > ... and this doesn't help with the runtime support that you need. > Here's a concrete example: that code will also have a GC in it, and > you'll end up running an emulated piece of code in JS to do the GC, > spending a good number of layers on work that JS already does. As I see it, I am getting the runtime without any effort at all :-) On the other hand it will be slow (but there are no free lunch). > In any case, there's nothing really special about LLVM in this context > other than the fact that someone wrote such a compiler. It doesn't > change much how "direct" things are, because it's at a low enough > level that there's not much difference. Just to make sure: I never meant to argue that an Emscripten Racket would be practical solution. It was more along the line of: Hey, here is a fun project idea. And my quick estimate of the effort needed was that it shouldn't take that long to get a demo running. Thus it is no catastrophe if it turns out to be too slow to use. That is I do not see it as a replacement of the current compiler. -- Jens Axel Søgaard _________________________________________________ For list-related administrative tasks: http://lists.racket-lang.org/listinfo/users
