From: Robert S. Thau (rst@ai.mit.edu)
Date: Tue Jan 16 2001 - 09:12:04 PST
Tony Finch writes:
> Stalin is a highly optimising
> compiler for Scheme that uses a loose version of type inference to
> approach the performance of ML. However to make the best use of it you
> have to write ML in Lisp syntax; good programmers do this anyway, but
> still they might as well just switch to ML and get a proper syntax and
> a compiler that tells them when they are being stupid.
Unfortunately, Stalin has another problem, which is compile time ---
easily several hours on a large scheme program (e.g., Stalin itself).
It's possible that this could be speeded up, but Siskind (Stalin's
author) evidently doesn't see that as a priority. There are people
actually using it for real code, but IIRC, they are using it mainly
for production, and use another scheme compiler for development.
The compile times of ML compilers, even the researchy ones, generally
allow for a sane edit-compile-debug cycle. (Though, as more than one
programmer has noted, in ML it's more like edit-typecheck-edit again;
once your code gets through the typechecker, it often Just Works.
What makes this particularly intriguing is that well-written ML code
contains very few type declarations; types are generally inferred from
context).
rst
This archive was generated by hypermail 2b29 : Tue Jan 16 2001 - 09:18:09 PST