[gclist] Re: gclist-digest V2 #77

[Thomas Lindgren]

I have added Mads Tofte to the recipients, so that he
might add some comments should he so wish.

>From the gc-list-digest:

At 10:45 AM 11/6/97 +0100, Christian Fabre wrote:
> ... what is "region inferencing"
>actually?
>
>A full dissertation is not necessary, a few lines would be fine.

You can get a working region inferencing system from:

http://www.diku.dk/research-groups/topps/activities/kit2/readme.html

It is an implementation of standard ML, with region inference that
handles higher-order functions and other goodies. Unfortunately,
the analysis as such is somewhat hairy, but I think closer
study may be worth the effort.

I saw a demo this spring; one nice aspect is that the max. required
memory is often quite small. Static bounds on memory use might be
possible from time to time.

There is also an interesting paper on how you write SML programs so
that the region inference will do the right thing. _That_ might (in
some suitably massaged form) well be the replacement of malloc(), if
I'm allowed a bold prediction.

	Best,
		Thomas
--
Thomas Lindgren, Uppsala University
e-mail: thomasl@csd.uu.se
http://www.csd.uu.se/~thomasl/