[cap-talk] Capabilities and pseudo-atomicity?
wil.pearson at gmail.com
Thu Feb 21 16:08:09 EST 2008
On 18/02/2008, Jonathan S. Shapiro <shap at eros-os.com> wrote:
> Question: If atomicity is desired, why isn't a perfectly conventional
> CAS operation sufficient?
I've been mulling this over. The answer I got was having the failure
occur at the correct place. I'm considering an atomic action as
meaning while one person is doing an action no one else has the
authority to do that action. Having the CAS* before the capability
invocation would not be practical, having it after would mean that the
caller would have to rely on the callee to inform them of the failure
or otherwise handle the failure. The right place for the error to
happen is in between the caller and the callee, as the caller doesn't
have authority at this point in time. So the caller can take whatever
the relevant steps would be to cope with the error.
How big a deal this is I am not sure, but it might be something. How
easy it will be to make objects that are actually atomic from it is
*By CAS I am meaning something like the old 68020 operator not the
More information about the cap-talk