ICFP?
Henry Cejtin
henry@sourcelight.com
Mon, 10 Sep 2001 15:17:36 -0500
If you are making more tee shirts, tell me how much they cost and send me 2
extras (one for Johann and one for Rico). They were truly a great idea.
I agree with what you said for the new opus, but with respect to the flow
analysis, I think that it is definitely worth talking about the optimizations
that make it fast enough that you can afford to use it. Certainly the stuff
about ignoring types that don't contain an arrow. I would say that also
something about the other hacks for this pass to. Perhaps even a statement
on safety (just to convince people that we know what we are doing) but no
proofs.
For the raise=>jump stuff, either in the new opus or a separate paper, it is
worth at least pointing out that this optimization triggers a lot more often
in MLton than it would in other ML compilers because of inlining. I realized
this when thinking about something Stephen said about using raise to get out
of fold loops early. Thanks to this optimization, and the fact that folds
are basically always inlined, this really is completly ok efficiency wise.
With regards to the handler stuff, we really need some more examples of
programs that actually use handle. count-graphs is a VERY heavy user, but it
would be nice to have some other, perhaps less ideo-syncratic ones.