Re: probably cluelessly beating a dead horse

>  >  Do you mean, given some capability, c, perform some operation that
>  >  returns c1 and c2, such that c1 and c2 must be used together to achieve
>  >  what could have been done with c?
>
>  totally, yup, right, yes.
>
>  But it would still need the ability to split the capability in half,
>  which brings us to your idea of using S/U to achieve that! :-) Cool.

I think I realized a few things more recently. When I think about
using S/U to make c1,c2 I realize that is what I think Dean had
described in the first place, effectively. So that got me thinking
that:

i) The complete capability must never go either unboxed or boxed
across the wire, because the unsealer will go across the wire, giving
the people along the route nefarious access.

ii) We can't use stuff like "random1 random2 random3 random4" because
that is a probabilistic argument which we're trying to avoid.

iii) I am assuming it is safe to assume that capabilities are unforgeable.



Therefore, I am currently thinking [to clarify what you had asked
about what i meant about that]:

i) we have to send c=(c1,c2) only as c1 c2 never as c.

ii) c1 and c2 can be capabilities and are individually unable to
transfer any of the money, and together can only transfer all of the
money e.g. c1="initiate the first half of the full transfer" and
c2="initiate the second half of the full transfer" (and presumably
they should be idempotent).

iii) we do not need any S/U pairs anywhere, we just send things plain
text since only KEQD will receive both c1 and c2 (since we can assume
Alice & Bob aren't working together).

?? :-}