Remove duplicates in frequency shrink tree

[TysonMN]

Fixes #319

We had the same bug that Haskell had. They fixed this bug in PR hedgehogqa/haskell-hedgehog#406. I tried to mimic their solution here. I think I was successful.

CC @HuwCampbell

I did one thing differently related to their shrink function. I think the input n of the lambda to shrink is put into the root of the shrink tree by shrink. Instead, I avoided an off-by-one error by first paring the weights and then calling takeWhile. Without pairing, the resulting sequence does not include the weight corresponding to the weight k in the base case of pick.

[TysonMN]

I tried to mimic their solution here. I think I was successful.

To be clear, I mean I think I was successful in fixing the bug. My implementation is probably not as elegant as @HuwCampbell's.

I force pushed the branch with some new commits that added another test. This test ensures that the entire shrink tree produced by frequency is balanced. Because a test failure means recursing into the corresponding subtree and a test success means moving on to the next child, a balanced shrink tree is one in which, at each subtree/Node, the number of children and the depth of that subtree differ by at most 1.

@HuwCampbell, although I was able to understand your Haskell code well enough, I don't understand it perfectly. Does your implementation produce balanced shrink trees?

[HuwCampbell]

No it didn't produce a balanced tree. We did it before our other research on balanced trees. I think it should be possible though.

[TysonMN]

Sure. The frequency shrink tree probably wasn't balanced back when then shrink trees included redundant values. What I meant was, is the shrink tree for a frequency generator in Haskell Hedgehog balanced (as defined above) now that shrinking happens without repetition?

[HuwCampbell]

No it's not.

It still uses the shrink function, so there's a small amount of duplication there.

[TysonMN]

Maybe that can be your next improvement then. I am still curious, after that change, if the resulting tree is balanced.

[ghost]

@TysonMN do you remember if this is ready to merge?

[TysonMN]

I think so. As I recall, the tests here are very strong.

[ghost]

Merged. 👍🏼