|
|
|
|
|
|
by YorkshireSeason
2855 days ago
|
|
|
I agree that formal verification in this problem space is currently not feasible (even if we ignore the oracle problem: where to get the spec from). But let's not make the perfect the enemy of the good. Even a basic reproducible test suite (e.g. running the software on some very basic pictures with a known result, and running against similar software) in an automated fashion, would help. Almost no scientific software I've seen get even those basics right. |
|
|