[cap-talk] Resource management on OCap systems

David Barbour dmbarbour at gmail.com
Tue Jan 3 14:46:20 PST 2012


On Tue, Jan 3, 2012 at 1:41 AM, Jed Donnelley <capability at webstart.com>wrote:
>
>  I agree that it's wise not to keep unneeded references.  I also agree
> that traditional (if by that you mean by reference count) GC isn't
> necessary.
>

Traditional GC also includes options such as:
* mark and sweep
* copy collection

There are also many GC models that are effectively hybrids with reference
counting, such as generational collectors or region-based collectors.


> Sorry, but I don't understand what all you mean by, "But by uniformly
> representing `connection` objects, we can also have linear references for
> relationships with shared resources."  What does uniformity have to do with
> things?
>

One problem with reference counting is that a `count` is a lossy
representation of an object's relationships with the world. Tracking
connections is more robust, and can simplify both GC design and application
semantics for disrupting or destroying the object. However, to generically
take advantage of this in the GC, it must be uniform - i.e. there is no way
to access any object except via connection.



When you discuss "linear references for relationships with shared
> resources" I'm reminded of the path based sorts of access that Alan Karp
> discusses with the Client Utility.  Do you know if you're referring to a
> similar concept?
>

I've not paid any attention to the discussion of Client Utility.

Regards,

Dave
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.eros-os.org/pipermail/cap-talk/attachments/20120103/6cf2b13c/attachment.html 


More information about the cap-talk mailing list