3 ms·
A few points resonated with me: Proving that the spec is the right spec is impossible (stealing a point from Roger Penrose). At some layer, you have a transit
by AnimalMuppet 8y ago
A few points resonated with me:
Proving that the spec is the right spec is impossible (stealing a point from Roger Penrose). At some layer, you have a transition from an informal idea in someone's head to a formal specification. Proving that the formal specification accurately matches the informal idea is impossible by formal means, since the idea being compared to is informal. Put another way, it is impossible to formally prove that the transition across the informal-to-formal boundary was done correctly.
Formal methods burn up a lot of time, which could be spent on other things - like iterating and refining the design and (informal) spec. If you build the wrong thing, formally proving that what you built matches the formal spec doesn't actually help. You might be better off getting a prototype in front of users/customers and getting feedback, rather than formally proving that you implemented your initial spec or design.
And, one thing the article doesn't mention: Formal verification isn't bulletproof. Quoting from memory: "It's amazing how many bugs there can be in a formally verified program!" (I believe said by Don Knuth.) Worse, the verifier can also have bugs.
Note well: This does not mean that formal verification is useless. It's just not 100% perfect. 95% is still worth something, though. If formal methods are worth doing, they're worth doing for only 95% of the benefit.