Jimmie Houchin <jhouchin at texoma.net> wrote:

> How does this compare to these discussions and the Squeak VM design?
> 
> The design of the Inferno virtual machine
> http://www.vitanuova.com/inferno/papers/hotchips.html
Well, almost the first thing they say about garbage collection is just
plain wrong; reference counting is _not_ exact, unless one has the
special case that reference counts can be functionally infinite. In any
practical case one will have potential for refCt field overflow and a
need to deal with it. Next, they completely miss the well documented use
of deferred reference counting, which (obviously) doesn't 'reclaim
memory the moment it becomes unused' but does cut refCt costs
enormously. They seem t be confusing generation scavenging with
mark-sweep; M-S doesn't need a large amount of extra memory (I assume
that is what they mean by 'a large arena'?) and even GS doesn't really
need a great deal normally.

Most of this paper is really trivial and seems unaware of at least
fifteen years of work in the field. The only intersting point is the
claim the 'Dis compiles quickly into native code that runs only 30-50%
slower than native C' though that is explained by the apparent similarty
of Dis to C anyway.

tim

-- 
Tim Rowledge, tim at sumeru.stanford.edu, http://sumeru.stanford.edu/tim
Useful random insult:- S  p  a  c  e  d   o  u  t .