Re: [Caml-list] Comparison of OCaml and MLton for numerics

[Brian Hurt <bhurt@janestcapital.com>]

Stephen Weeks wrote:

> The same arguments that one can make for simple compilation
> strategies leading to predictable performance could be used to argue 
> that we
> should all program in C or in assembly.


What's humorous about this statement is that most C compilers now are 
implementing a lot of these fancy optimizations, even whole-program 
optimizations (see LLVM), which makes predicting their performance 
equally tricky to predict. 

I think the real reason Ocaml doesn't have advanced optimizations and 
whole program analysis is just one of time vr.s value.  It hasn't been 
valuable enough to someone yet to take the time and put in the work to 
implement them.  The position of most people on this list, including 
INRIA, seems to be "it'd be nice, and we'd definately use it if it were 
available, but at the moment we're doing something more important/more 
interesting/else..."

Brian