I know of two approaches to property checking:
-
Randomized approach (as in QuickCheck) makes you define a generator of random values for your types, then verifies your invariants for each of a large number of randomly generated cases. For example, in case of a vector space ℤⁿ (defined, say, as
[Word]
), it would generate vectors of any length and direction. -
Deterministic approach (as in SmallCheck) works the same, except that it generates valus from simple and small to more complex and extensive deterministically, covering (as I understand) a small part of domain tightly. For example, in the same case as above, it would generate zero vector, then all vectors of length <= 1, then all vectors of length <= 2, and so on, eventually covering all interesting values (such as unit vectors), for which there may conceivably be unforeseen corner cases that break an invariant.
Did I get it correct? What are the benefits and downsides of either approach? What are the preferred use cases for either? Maybe, for best results, both approaches should be combined, say, covering some moderately sized "base" with a deterministic checker and then a random sample with a randomized one?
I'm looking for some solid, motivated best practice to use daily and advance from now on.
Aucun commentaire:
Enregistrer un commentaire