Testing Shoki

[jnape]

I'd like to leverage property testing for the data structures in shoki, and I'm going to try plugging gauntlet in for that. If it's alright with you, I'd like to keep this issue open so I have a place to ask questions. Do you have any anticipation of when you might have a GA release?

[kschuetz]

I think the API is starting to stabilize. (Several implementation details, such as shrinking, still need work).

For now you need to instantiate the default API with the gauntlet() call (then either use that, or override settings with an immutable builder).

I'm open to any suggestions or comments!

Here is a simple example:

gauntlet/gauntlet-core/src/test/java/dev/marksman/gauntlet/examples/VectorTests.java

Line 16 in cb57bd7

void reverseTwiceIsOriginal() {

[jnape]

After about 30m of use, so far my principle suggestion is to not abandon this project. Writing an Arbitrary<shoki.HashMap<K,V>> given an Arbitrary<K> and an Arbitrary<V> was straightforward (and even fun!) after understanding the model. My feeling is that you're on to something good with gauntlet.

Any plans for a JUnit Reporter? I imagine that would interface seamlessly with IntelliJ, which would be wonderful.

Seriously - from everything I've seen, my quibbles are mostly over trivialities. Nice job.

[kschuetz]

Great! Thanks for the feedback. I will move forward with it. Let me know if you come up with any new uses cases in shoki.

I do have plans to improve the Reporter later on (right now, it's essentially a spike).

[jnape]

This was sublime:

@Test
public void merging() {
    GeneratorTestApi<Tuple2<HashMap<Integer, String>, HashMap<Integer, String>>> harness = gauntlet
            .all(arbitraryHashMap(arbitraryInt(), arbitrary(generateString())).pair())
            .withSampleCount(_100);

    harness.mustSatisfy(prop("∀(k,v) ∈ m1.merge(m2, sg) : just(v) == present(sg, m1.get(k), m2.get(k))",
                             t -> t.into((m1, m2) -> {
                                 Semigroup<String> sg = join();
                                 return and()
                                         .foldMap(into((k, v) -> eq(just(v), present(sg, m1.get(k), m2.get(k)))),
                                                  m1.merge(m2, sg));
                             })));
}

FYI that t -> t.into cannot be replaced with into because prop does not accept an Fn1 that supports contravariantly positioned inputs. However, if prop allowed Fn1<? super A, Boolean>, that would allow the better ergonomics, if you're open to it.

[kschuetz]

Nice! I will look into making that change.

[kschuetz]

This is gradually coming together. Beginning to do some dogfooding to get a feel for how comfortable the API is. Any suggestions are welcome!