[MLton] Multicore CPU's and MLton

Stephen Weeks MLton@mlton.org
Tue, 5 Jul 2005 10:14:46 -0700


> Oh, I guess this got lost in the discussion, but I don't really care how 
> parallelism is implemented. Process, level ala parallel make would be 
> fine for me.  I just want to make sure that MLton can use that extra 
> core for something!

Most of the discussion was about MLton-generated executables taking
advantage of multi-core.  As to MLton compiles, all you're gonna get
is the codegen stuff we already talked about, and process-level
parallelism is enough for that.

> I'm concerned that the whole program methodology while quite elgeant
> and practical today, will look less and less so in the future.

I completely disagree.  Whole-program compilation in MLton has never
been about providing rapid turnaround to be the sole development
solution.  It's about providing performance and eliminating the costs
of abstraction so that developers can write clean code.  All that is
necessary (and which we have achieved) is short enough compile times
that the whole-program compiler can be used frequently enough and
without too much impedance of the overall development process.

One must still have a non-whole-program compiler for development.
Separate compilation is a much better fix for this problem than
parallelized whole-program compilation.