[cap-talk] use of hashcodes?

[David Wagner]

Raoul Duke  wrote:
>On Fri, Feb 19, 2010 at 6:49 PM, David Wagner <daw at cs.berkeley.edu> wrote:
>> As others have said, it's not really a matter of probability [1].
>
>my thought was less about cracking and more about random dumb luck
>debugging -- hey, suddenly we got a really weird response for our
>request because hashes collided. of course unlikely. supposedly more
>likely for one to get random hardware bugs instead, which is more of a
>perspective i can grok.

Yes.  It's more likely that you get struck by lightning three
times in a row (that actually might happen, whereas running across
a collision in SHA256 by pure chance: no way).

>given that i'm not thinking about somebody cracking, i'm thinking
>about random dumb bad luck, i just wonder why we pick probabilistic
>things when we really want something w/out collisions.

Because we have no choice: the pigeonhole principle says that
collisions will exist any time the hash output is shorter than
the hash input.  All we can do is make it exceedingly unlikely
that we'll ever run into those collisions in practice: we cannot
eliminate them entirely.