[MLton] Growing the user base...

Daniel C. Wang danwang@CS.Princeton.EDU
Tue, 03 Aug 2004 12:50:21 -0400


Stephen Weeks wrote:

>>BTW.. I'm teaching the PL course at Princeton this year and would
>>have liked to use MLTon because of the FFI. In particular I wanted
>>to do a cool project with OpenGL and MLTon. MLton's FFI seems less
>>of a hassle then SML/NJ in this respect.
> 
> 
> Sounds great, and I agree with everything Matthew said.  I'm only
> worried about the "would have" part :-).  Have you decided not to use
> it?

I decided not too, because of the following reasons... Ullman's Elements of 
ML uses SML/NJ as the reference system. I'm going to stick with 110.0.7 to 
avoid any big gap between the examples in the text and the actual system.
So there is legacy documentation that I want to be compatible with to avoid 
any hassles. I'm also a bit concerned about resource usage. These aren't big 
programs that people are going to be compiling, but there maybe 10+ students 
  banging on on poor server with MLton installed.

A good interactive interpreter that mimics the behavior of SML/NJs toplevel 
(or improves on it!) would probably address most if not all of my complaints.

My ideal MLton system would look something like the following. A byte-code 
compiler/interpter that compiles to byte-code and supports separate 
compilation. An option to compile a bunch of byte-code files into one 
highly-optimized standalone executable.

Extra credit of clean interoperatoin betweeing byte-code and natively 
compiled ML code.