3 ms·
Re-read this article every time when you come across yet another blog post or article extolling the virtues of formall proofs, verifications, and, while we’re a
by dmitriid 9y ago
Re-read this article every time when you come across yet another blog post or article extolling the virtues of formall proofs, verifications, and, while we’re at it, static typing.
None of these work when taken in isolation and given to actual people to work with.
- TeMPOraL 9y agoOn mobile and imgur refuses to work on my Fx, so won't post a screenshot. So just imagine how funny yours and oconnore's [0] comments look next to each other... So, which way is it? [0] - https://news.ycombinator.com/item?id=15488391 https://news.ycombinator.com/item?id=15488391
- ktta 9y agoIf you look around, the problem touted here was that formal verification was done, but independently for two different parts of the standard, in isolation. The layman answer that is formal verification should've been done on those systems after integration. I'm not sure how much that would've helped though, since the proofs for each of those was complex enough. And then actual code written tends to sway far away from what protocol implementers (also crypto folks) think it should look like. So I'd say more formal verification, but also make it easier and accessible for normal developers without a Math PhD. And verification of actual code would be ideal. Something I've come across recently is formal verification of cryptographic implementations not on paper (i.e. mathematically) but in plain C using tools and languages developed recently. For example, a three part series on how SAW and Crytol were used to formally verify s2n (AWS's TLS lib): https://galois.com/blog/2016/09/verifying-s2n-hmac-with-saw/ https://galois.com/blog/2016/09/verifying-s2n-hmac-with-saw/ https://galois.com/blog/2016/09/specifying-hmac-in-cryptol/ https://galois.com/blog/2016/09/specifying-hmac-in-cryptol/ https://galois.com/blog/2016/09/proving-program-equivalence-with-saw/ https://galois.com/blog/2016/09/proving-program-equivalence-... PS(A) for the adventurous: I've also come across a github repo that lists related stuff https://github.com/enzet/symbolic-execution https://github.com/enzet/symbolic-execution
- dmitriid 9y agoThe protocol was formally verified. And yet...
- tensor 9y agoYet all of these are about 100x better than the cowboy attitude 99% of developers these days have. So I'd say that every time you come across a blog post extolling the virtues of formal proofs, verifications, and static typing, you should listen. Because you are awful at programming, and security, and these things will help you. Sure, they won't be enough, but to think they are anything other than minimum requirements for security is being incredibly arrogant and deluded.
- dmitriid 9y agoI totally agree with you. There are just too many articles and blog posts that read like "formal proofs and/or static types are end of all need for anything like tests be cause while the program is verified and/or compiled, it's correct". As programmers we are always somewhere in the middle :)
- platz 9y agoDid you skip that last 4 paragraphs?
- dmitriid 9y agoIt's precisely because I didn't skip the last four paragraphs that I wrote my comment. Here's how the fourth last paragraph starts: > The critical problem is that while people looked closely at the two components — handshake and encryption protocol — in isolation, apparently nobody looked closely at the two components as they were connected together.