Intersection is too slow for case class with more than 6 fields

[sergei-grigorev]

So, shapeless.ops.hlist.Intersection takes too much time to compile. In my case it's impossible to wait for a case class with 10 fields. But for a test only - 6 and 7 fields also takes too much time.

    test("too slow to compile") {
      case class C1(f1: String, f2: String, f3: String, f4: String, f5: String, f6: String, f7: String)
      case class C2(f1: String, f2: String, f3: String, f4: String, f5: String, f6: String, f7: String)

      val gen = LabelledGeneric[C1]
      val gen2 = LabelledGeneric[C2]
      val inter = hlist.Intersection[gen.Repr, gen2.Repr]
    }

In my case of course classes is a little different, it's just a test case. Is there any compiler optimization to pass this problem?

[milessabin]

Have you tried TLS with -Yinduction-heuristics?

[sergei-grigorev]

@milessabin have tried but the situation almost isn't changed, it still takes ages. For example, 37 seconds for 7 fields and 108 seconds for 8 fields. I haven't got result for 9 fields

[joroKr21]

@sergeygrigorev a lot of fields is not enough to make it slow. Your example above compiles in a few seconds. I think the worst case would be to have a lot of fields of different types in reverse order. Like so:

case class C1(f1: Byte, f2: Short, f3: Char, f4: Int, f5: Long, f6: Float, f7: Double, f8: String)
case class C2(f1: String, f2: Double, f3: Float, f4: Long, f5: Int, f6: Char, f7: Short, f8: Byte)

The problem with type level programming is that you can do mostly naive algorithms and often end up with quadratic worst case complexity.

[Ssstlis]

Should we close this issue as PR as well?

[joroKr21]

Fixed by #682