Infer dependent parameter in NonEmpty/ParallelTests/Laws #3046
Conversation
Codecov Report
@@ Coverage Diff @@
## master #3046 +/- ##
==========================================
+ Coverage 93.46% 93.46% +<.01%
==========================================
Files 368 368
Lines 6975 6979 +4
Branches 187 184 -3
==========================================
+ Hits 6519 6523 +4
Misses 456 456
Continue to review full report at Codecov.
|
|
Random clarification: this is only source-breaking if someone wrote their own laws derived from |
There was a problem hiding this comment.
I didn't make this change in #3012 intentionally, since I thought it would be better to make what's tested more explicit, and since writing out the extra type parameter once isn't really a big usability issue in tests. I meant to flag that decision in the PR but forgot.
Given that this PR has turned up issues downstream I now think it's probably the right thing to do.
| def apply[M[_]](implicit ev: NonEmptyParallel[M]): NonEmptyParallelLaws.Aux[M, ev.F] = | ||
| apply[M, ev.F](ev, implicitly) | ||
|
|
||
| def apply[M[_], F[_]](implicit ev: NonEmptyParallel.Aux[M, F], D: DummyImplicit): NonEmptyParallelLaws.Aux[M, F] = |
There was a problem hiding this comment.
It doesn't really matter but I did this in the other direction in #3031, with the DummyImplicit on the new one.
It's a little wonky. Also I left one
.Auxusage in the tests which came from the fact that the compiler just wasn't able to infer the dependent version. I didn't look into why, but I suspect it's wonkiness withOneAndand a scalac bug. Putting up the @milessabin signal!Also this introduces an interesting pattern, which is adding
def Auxalongsidetype Auxwhen you havedef apply. I would propose we should do this forParallelitself as well, but I didn't make that change. I think this is better than the currentDummyImplicitapproach, but I'm okay with going withDummyImplicitinstead.