Make Parallel's F a type member

[travisbrown]

Fixes #2233. We'd been holding off on this change because we thought it would necessarily require both source and binary compatibility breakage, but it turns out we can fix it without breaking binary compatibility with Cats 1.x.

Note that this change does break source compatibility pretty dramatically, but the migration path will generally be pretty clear (add .Aux, maybe an explicit type parameter here or there). Anyway, as Rob Norris says here, the current API is so awkward to use that this is unlikely to break much code.

With this change you can use parallel syntax without having to thread an extra type parameter everywhere through your code. Taking an example from the README of cats-par (a library for working around the limitations of the current API), this PR allows us to write the following:

import cats._
import cats.implicits._
import cats.data._

def withNewParallel[F[_]: Monad: Parallel, A, C, D](
  as: List[A],
  f: A => Kleisli[F, C, D]
): Kleisli[F, C, List[D]] =
  as.parTraverse(f)

With the current API we have to include an extra parameter G[_], both here and all the way up our call stack:

def withOldParallel[F[_]: Monad, G[_], A, C, D](
  as: List[A],
  f: A => Kleisli[F, C, D]
)(implicit P: Parallel[F, G]): Kleisli[F, C, List[D]] =
  as.parTraverse(f)

Note that the only changes to the tests are a few Auxs (and one fewer type parameter in one applicativeError call); otherwise the usage exercised in the tests didn't require changes.

[LukaJCB]

Looks good to me, thanks! This has been a long time coming

[codecov-io]

Codecov Report

Merging #3012 into master will decrease coverage by <.01%.
The diff coverage is 98.11%.

@@            Coverage Diff            @@
##           master   #3012      +/-   ##
=========================================
- Coverage    93.6%   93.6%   -0.01%     
=========================================
  Files         368     368              
  Lines        6948    6946       -2     
  Branches      188     185       -3     
=========================================
- Hits         6504    6502       -2     
  Misses        444     444

Impacted Files	Coverage Δ
...ain/scala/cats/laws/discipline/ParallelTests.scala	100% <ø> (ø)	⬆️
...a/cats/laws/discipline/NonEmptyParallelTests.scala	100% <ø> (ø)	⬆️
core/src/main/scala/cats/syntax/parallel.scala	62.5% <0%> (ø)	⬆️
core/src/main/scala/cats/data/NonEmptyList.scala	100% <100%> (ø)	⬆️
core/src/main/scala/cats/Parallel.scala	84.37% <100%> (ø)	⬆️
core/src/main/scala/cats/package.scala	100% <100%> (ø)	⬆️
core/src/main/scala/cats/data/WriterT.scala	91.26% <100%> (-0.07%)	⬇️
...rc/main/scala/cats/laws/NonEmptyParallelLaws.scala	100% <100%> (ø)	⬆️
core/src/main/scala/cats/data/Ior.scala	98.52% <100%> (ø)	⬆️
core/src/main/scala/cats/data/Kleisli.scala	90.9% <100%> (-0.09%)	⬇️
... and 5 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 14e525a...5d7ab29. Read the comment docs.

[kailuowang]

NIce, how come this (it doesn't break BC) didn't occur to us.

[kubukoz]

y'all reading in my mind, I thought about this just a couple days ago... a very welcome change!

[SethTisue]

recent discussion at scala/scala#9303 has us wondering if this change was in fact safe. (which might affect decisions to make other such changes in the future)

[kubukoz]

Wow, took us long enough!

I can't remember any case in the ecosystem where this broke anything, but I would wager Parallel isn't that popular an abstraction in libraries, which could be why - everyone just updated cats and their sources and they were fine. (I think)

[SethTisue]

Perhaps y'all should mimaReportSignatureProblems := true and see whether there are too many false positives.