2 ms·
> can be attacked on that message until its stack dumps This sounds like an attempt to combine intuitions about the pumping lemma with memory corruption vulner
by schoen 10y ago
> can be attacked on that message until its stack dumps
This sounds like an attempt to combine intuitions about the pumping lemma with memory corruption vulnerabilities: that "any system" must have some inputs that are "too much" for it to interpret correctly (for many programs written in C, that could be a single input string that's "too long").
https://en.wikipedia.org/wiki/Pumping_lemma https://en.wikipedia.org/wiki/Pumping_lemma
https://en.wikipedia.org/wiki/Buffer_overflow https://en.wikipedia.org/wiki/Buffer_overflow
These intuitions are founded on good reasons because every computing system has different kinds of limitations, and some kinds of distinctions that it can't make correctly, and some kinds of failure modes if it's expected to make arbitrary distinctions.
But there are ways out:
* If we narrow the definition of the task that the system has to do, it may be possible for the system to do it perfectly, and it may even be possible to make a formal proof that the system performs that task perfectly, even though there are other more general tasks that it doesn't perform, or performs imperfectly.
* If we allow the system to "give up" or indicate that it refuses or fails to answer certain questions, then it doesn't have to be "wrong" or exhibit wrong or unpredictable behavior. It can just say "sorry, not sure", which may be annoying but isn't a mistake and isn't necessarily a security vulnerability. (Maybe this is akin to those movie robots saying "does not compute" all the time.)
Within these frameworks, a kind of perfection, or at least a kind of correctness, is possible for many computing tasks -- including some that involve Turing-completeness, either as a property of the environment they're implemented in or as a property of the machine they implement.