4 ms·
A while back on here, I was arguing against common methods of programming for security-critical software in favor of high assurance methods. Those methods used
by nickpsecurity 11y ago
A while back on here, I was arguing against common methods of programming for security-critical software in favor of high assurance methods. Those methods used mathematical specification, refinement, analysis, and so on to know everything the program could do in success or failure. The implementation was similarly provable due to how it was implemented: state machines. All high assurance that I could recall used interacting state machines (ISM's or ASM's) with modularity, abstraction, and careful interfaces. I also remember being surprised that "kragen" correctly guessed FSM's as what I was advocating.
State machines are very versatile. One can do provably safe transformations on them with suitable tools. One can derive them from functional languages where most of developing and proving work is done. I've seen several techniques of high-level synthesis for hardware that produce, split, optimize, merge, etc various FSM's to go from high-level function to logic gates. Many formal verification tools, esp easier model checkers, work well with them. All this makes them a powerful tool worth mastering and using more than we do.
- nextos 11y agoSo how would you rank model checking, abstract interpretation and theorem proving according to your experience in terms of practicality?
- nickpsecurity 11y agoAbstract interpretation (esp Astree Analyzer) at top for ease of use in static analysis. Fire and forget mostly. Model checking next because you just learn specifications and certain ways of structuring code for easy analysis. Theorem proving last for where it's worth the specialist labor and long turnaround (eg kernels, compilers, SSL). That's been my order for years. Empirical evidence supports it so far.