On Monday 08 December 2008 23:52:02 Jon Harrop wrote:
> I think the interesting question is "how might we accomplish this"...

I have been giving this question a lot of thought recently. The problem is 
that the goal is quite obvious and objective but the route is not at all 
obvious.

I think the best way to proceed is to build a DSL that is initially useful 
(e.g. for numerics) and use macros and LLVM to get it JIT compiled. Then grow 
the language and garner users until it is feasible to create a stand-alone 
version of the language.

I believe it is quite feasible to create a MiniML with its source quoted 
inside OCaml programs using Camlp4 and an LLVM backend for it that beats 
OCaml's performance on some tasks. That alone would be useful and it can be 
improved incrementally from there until it is a full-blown language 
implementation.

-- 
Dr Jon Harrop, Flying Frog Consultancy Ltd.
http://www.ffconsultancy.com/?e