4 ms·
I disagree. Good formal systems allow you to specify global correctness conditions (e.g., "the database is always consistent", "data is never lost if a machine
by pron 9y ago
I disagree. Good formal systems allow you to specify global correctness conditions (e.g., "the database is always consistent", "data is never lost if a machine fails") in a concise and clear way (a couple of lines). The correctness of that specification is then relatively easy to verify by inspection. You then check that the system as a whole satisfies those requirements.
It's true that to date we don't have tools that can efficiently verify a program all the way from code to the high-level global correctness conditions (and, TBH, I don't think we'll have them in the foreseeable future, if ever), but there are tools that help narrow the gap: code-level tools that work at the code unit level (function preconditions), and high-level tools that work at the large-scale system description. Using both (which is exactly what companies that use formal methods for software, like Altran UK, do), let you work from both directions -- top and bottom -- to minimize the "uncertainty" gap, and, most importantly, they do it affordably.
It's also true that formal methods are a tool that we've not yet learned to effectively use in practice, and don't yet know the best ways to deploy it. But we're learning, and we're already at the point that at least some are seen considerable payoffs.