3 ms·
Unless they are executable and part of the language tooling like SPARK, Dafny and co, I see little value coding an abstract model, that might be validated in th
by pjmlp 1y ago
Unless they are executable and part of the language tooling like SPARK, Dafny and co, I see little value coding an abstract model, that might be validated in theory, but hardly anyone will be able to cross-check if the manually written language ZYX code actually maps to the validated model.
And then there are the code changes, which most likely devs aren't going back to the formal model to validate them, before applying the new set of changes.
- dataflow 1y agoI agree with you on the first paragraph, but the second one is something you could presumably enforce with tooling and during reviews, no?
- pjmlp 1y agoIn theory, in practice even enforcing unit tests is a challenge, and most code reviews end up being about code style than anything else, if they happen.
- bbkane 1y agoAWS recently mentioned something called PObserve in an ACM article. They said it checks logs emitted from the running application against the formal model. Sounds really useful for this exact problem; hopefully one day they'll open source it.
- pjmlp 1y agoI would need to check the article, although it sounds very fragile, expecting some logs are present in a specific format, and they actually tell the truth about what the code is doing.