Revert "Fix #3348: use a new typeState in inferView"

[liufengyun]

Reverts #3354

@odersky : I'm not sure exactly what you proposed here. Could you elaborate more or take this over?

[allanrenucci]

My understanding was to try this instead:

-     if (ctx.mode.is(Mode.ImplicitsEnabled))
+     if (ctx.mode.is(Mode.ImplicitsEnabled)) {
+       val savedConstraint = ctx.typerState.constraint
        inferView(tree, pt) match {
          case SearchSuccess(inferred, _, _, _) =>
            adapt(inferred, pt)(ctx.retractMode(Mode.ImplicitsEnabled))
          case failure: SearchFailure =>
+           ctx.typerState.constraint = savedConstraint
            if (pt.isInstanceOf[ProtoType] && !failure.isInstanceOf[AmbiguousImplicits])
              // don't report the failure but return the tree unchanged. This
              // wil cause a failure at the next level out, which usually gives
              // a better error message.
              tree
            else recover(failure)
        }
+     }

[liufengyun]

Thanks @allanrenucci . Unfortunately this doesn't work, as the real problem is the shared typerstate, saving the constraint doesn't help here.

[odersky]

test performance please

[dottybot]

performance test scheduled: 1 job(s) in queue, 0 running.

[odersky]

@liufengyun I think we should just drop the assertion in implicitSearch. I don't see a reason after all why attempted completions should be undone, so we have to admit that they can influence the constraints.

[dottybot]

Performance test finished successfully:

Visit http://dotty-bench.epfl.ch/3370/ to see the changes.

Benchmarks is based on merging with master (e1365d9)

[allanrenucci]

I you keep working on this, can you reopen from dotty-staging as every push to lampepfl/dotty triggers a build

[liufengyun]

Superseded by #3374.