I don't think this is intended to be able to replace LLVM. Julia could be implemented on top of PyPy/RPython, but PyPy itself has an LLVM backend (one of multiple backends I believe). LLVM is a full toolchain for handling code --- generation, optimization, linking, analysis, and debugging. We need all of those capabilities, which go a good deal beyond just running julia code fast.
On Tue, Oct 27, 2015 at 5:30 PM, cdm <[email protected]> wrote: > > but it has its own virtual machine ... > which could conceivably replace > the LLVM ties ... ? > > ... unless that is inconceivable. > > > > i am glad that no one has rejected > the idea out of hand; further proof > that the Julia community is truly > remarkable. > > > > On Tuesday, October 27, 2015 at 7:31:52 AM UTC-7, Jeff Bezanson wrote: >> >> Looks very cool! Of course I'd be interested in compiling julia's >> frontend with something like this. We could badly use the extra speed. >> I assume it supports macros, in which case I hope somebody makes a >> library to undo the silly identifier renaming (e.g. defn -> define) so >> it can run more scheme-like code. >> >> This looks like a much more serious project than femtolisp, but I >> would like to dispute the claim that it is "lightweight". From the >> readme, it depends on python/pypy, libffi, libedit, libuv, and >> libboost, and is 10MB which makes it literally 100 times larger than >> femtolisp.
