[cap-talk] Reference count based garbage collection seen as flawed

Norman Hardy norm at cap-lore.com
Sat Jan 7 18:51:51 PST 2012

On 2012 Jan 7, at 17:35 , David Barbour wrote:

> On Sat, Jan 7, 2012 at 5:24 PM, Norman Hardy <norm at cap-lore.com> wrote:
> Point number 2 is otherwise contradicted by the fact that Keykos ‘persisted’ objets with reference cycles across checkpoint restarts.
> Does KeyKOS notify immediately when the last reference to an object is eliminated? Keep in mind that Shap's list regarded the architectural implications of immediate notification in the context of persistence. 
> Regards,

No Keykos does not signal unreachability.
I had lost track of the notification requirement since we considered that it was often necessary to delete objects despite existing references to them.
Such bloody stumps are efficiently guarded against by code using them, but the cost of surviving a missing limb may still be large, or impossible.

In general a designer guards against encountering missing 'constituent parts' by relying on the provider of those parts which generally requires passing a storage access capability (space bank) during the request for the new object.
Disappearance of 'exogenous' objects must generally be explicitly tolerated.
This is efficient but perhaps complex.
Those patterns are the center of this thread however. Sorry.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.eros-os.org/pipermail/cap-talk/attachments/20120107/9a4d4f95/attachment.html 

More information about the cap-talk mailing list