Comprehensive rewrite of List tests

[jonathanhefner]

I rewrote the List tests to be more comprehensive. The tests now cover all exposed API functions, with the exception of map3..5 (which were also previously not covered). I can add coverage for map3..5 as well as non-exposed functions, if desired.

The tests are also now parameterized, so that empty list, 1-element list, and N-element list use-cases are all covered equally. This also enables easy stress testing, either for stack overflow checking or for benchmarking.

I tried to keep the code as clean and clear as possible. If there are any parts that need improvement, please let me know.

[evancz]

Very cool, thank you!

My main concern is that if reverse is broken, lots of tests will behave in odd ways. That feels weird to me.

[evancz]

Very cool, thank you!

My main concern is that if reverse is broken, lots of tests will behave in odd ways. That feels weird to me.

[jvoigtlaender]

Regarding the reverse concern, wouldn't the obvious solution be to replace the reverse imported from List by a local, guaranteed-to-be-correct implementation of reverse in these lines:

https://github.com/elm-lang/core/blob/master/tests/Test/List.elm#L111
https://github.com/elm-lang/core/blob/master/tests/Test/List.elm#L157
https://github.com/elm-lang/core/blob/master/tests/Test/List.elm#L161-162
https://github.com/elm-lang/core/blob/master/tests/Test/List.elm#L166-167

?

[jvoigtlaender]

Regarding the reverse concern, wouldn't the obvious solution be to replace the reverse imported from List by a local, guaranteed-to-be-correct implementation of reverse in these lines:

https://github.com/elm-lang/core/blob/master/tests/Test/List.elm#L111
https://github.com/elm-lang/core/blob/master/tests/Test/List.elm#L157
https://github.com/elm-lang/core/blob/master/tests/Test/List.elm#L161-162
https://github.com/elm-lang/core/blob/master/tests/Test/List.elm#L166-167

?

[jonathanhefner]

I agree, it was a concern of mine. The problem is (without TCO) you have to rely on either reverse or map (or their underpinnings: foldl and foldr, respectively). Basically, there is always some non-primitive function you are relying on. But it may be worthwhile to remove as many layers as possible, e.g. limit yourself to just foldl, so I'll make another pass at it soon.

[jonathanhefner]

I agree, it was a concern of mine. The problem is (without TCO) you have to rely on either reverse or map (or their underpinnings: foldl and foldr, respectively). Basically, there is always some non-primitive function you are relying on. But it may be worthwhile to remove as many layers as possible, e.g. limit yourself to just foldl, so I'll make another pass at it soon.