[cap-talk] Access Abstraction vs Static Typing (was: David Wagner's Google techtalk is now up!)

[Jonathan S. Shapiro]

On Thu, 2007-12-13 at 12:11 -0800, Mark Miller wrote:
> Ok, now I think I mostly understand the question. Except that the
> question I think I understand would convert from a received byte
> sequence to an typed object by safe unserialization, not upcasting.

Yes conceptually, but the observation that I came to with Scott Smith
was that you can sometimes (in fact, often) do safe unserialization by
validating the range-constrained scalars in place on an immutable byte
sequence, and then return a deeply immutable pointer with the upcasted
type that aliases the same storage. By doing this you avoid all of the
copying overheads that typically are associated with deserialization.

This is why the prospect of using a similar trick for reply message
demultiplexing might be interesting.

shap