Fix how implicit candidates are combined

[dwijnand]

No description provided.

[odersky]

I think this is getting a bit too complex for comfort.

One alternative approach to consider is to have a smarter merge where we don't immediately designate a shadowed set, but merge outerImplicits and innerImplicits according to what is better:

• high levels beat lower
• at equal level, definitions beat imports and named imports beat wildcard imports

[odersky]

Something like this:

    private def combineEligibles(ownEligible: List[Candidate], outer: ContextualImplicits): List[Candidate] =
      val outerEligible = outer.uncachedEligible(tp)
      if ownEligible.isEmpty then outerEligible
      else if outerEligible.isEmpty then ownEligible
      else
        val ownNames = mutable.Set(ownEligible.map(_.ref.implicitName)*)
        if level == outer.level then
          val keptOuters = outerEligible.filterConserve: cand =>
            if ownNames.contains(cand.ref.implicitName) then
              val keepOuter = cand.level == level && cand.bindingPrec.id > bindingPrec.id
              if keepOuter then ownNames -= cand.ref.implicitName
              keepOuter
            else false
          val keptOwn = ownEligible.filterConserve: cand =>
            ownNames.contains(cand.ref.implicitName)
          keptOwn ::: keptOuters
        else
          ownEligible ::: outerEligible.filterConserve: cand =>
            !ownNames.contains(cand.ref.implicitName)

I assume a field bindingPrec: Typer.BindingPrec in ContextualImplicits and Candidate.

Not sure what to do under 3.0-migration.

[odersky]

LGTM now