Re: [ats-lang-users] The future

[Hongwei Xi <hw...@cs...>]

On Wed, 19 Nov 2008, Jon Harrop wrote:

>>Our company specializes in the use of functional programming languages 
>>(primarily OCaml, F# and Mathematica) for scientific computing (primarily 
>>interactive technical computing). I have written books on OCaml and F# and we 
>>sell a variety of software products. I have also evaluated several related 
>>languages (Haskell, SML, Scala, Clojure, Lisp, Scheme, etc.) but found them 
>>to be unsuitable for our needs for various reasons.
>>
>>OCaml provides a hugely productive and expressive foundation for technical 
>>computing that is popular among academics from many technical backgrounds. F# 
>>takes the core of OCaml's benefits and adds industrial-strength libraries and 
>>the ability to commercialize code.

The previous compiler for ATS (ATS/Geizella) was actually written in 
OCaml. Without OCaml, ATS would probably not even be here today. That 
being said, ATS is really designed to do things like systems programming, 
which do not seem to be Ocaml's cup of tea.

In terms of time/space performace, which we can readily measure, I am
very comfortable with the idea of ATS being played against OCaml. In terms 
of safety guarantees, the type system of ATS is evidently a lot more 
expressive than that of OCaml. The downside of ATS is that it may be
much harder to learn than OCaml, largely due to the richness and novelty
of ATS and the lack of documentation (and other facilitating tools).

>>I believe such applications would benefit from two main features that (AFAIK) 
>>have not yet been implemented in ATS:
>>
>>. A performant interactive top-level.
>>
>>. A safe, platform-independent and commerce-friendly intermediate 
>>representation for DLLs.
>>
>>The most obvious solution to me is an LLVM-based backend that might use CLang 
>>to JIT compile the output of the current compiler (assuming that is 
>>feasible).
>>
>>Once that infrastructure is in place I would be interested in developing an 
>>open source Mathematica-like IDE for ATS built upon OpenGL that users could 
>>buy and sell libraries for (and we can sell books about!).

I was actually thinking about generating LLVM code two weeks ago. This
doesn't seem very hard. But right now, we are really short of man power.
We have to stay focused on systems programming. I will keep this as a
possible project for the future.

--Hongwei Xi

Computer Science Department
Boston University
111 Cummington Street
Boston, MA 02215

Email: hw...@cs...
Url: http://www.cs.bu.edu/~hwxi
Tel: +1 617 358 2511 (office)
Fax: +1 617 353 6457 (department)