Imported From: https://issues.scala-lang.org/browse/SI-7212?orig=1
Reporter: @paulp
See #7873

@adriaanm said:

Paul Phillips
to scala-internals

It's interesting that you have some choices when looking for an unhelpful type to be inferred. Never let it be said the compiler is autocratic. You can have any color you want, as long as it's black or ebony.

trait A { def f: List[String] }
trait B extends A { def f = Nil }            // infers Nil.type
trait C extends A { def f = List() }         // infers List[Nothing]
trait D extends A { def f = List[String]() } // thanks for the type inference

// Why you might care
trait B1 extends B { override def f = List("a") }
trait C1 extends C { override def f = List("a") }
// ./a.scala:6: error: type mismatch;
//  found   : List[String]
//  required: scala.collection.immutable.Nil.type
// trait B1 extends B { override def f = List("a") }
//                                           ^
// ./a.scala:7: error: type mismatch;
//  found   : String("a")
//  required: Nothing
// trait C1 extends C { override def f = List("a") }
//                                            ^

@adriaanm said:
This likely makes incremental compilation juuuust a little bit harder still.

@adriaanm said:
Wish I had gotten around to do this -- will have to wait until 2.12, but thinking of enabling it under -future in 2.11.1.

@adriaanm said:
Design discussions get a wider audience on scala-internals, so I recommend discussing there. The inherited return type seems more appropriate to me

@adriaanm said:
I've made a bit of progress on this while reworking our fields encoding, but this will have to go in under a flag first during 2.12.x.

@som-snytt said:
I just did that thing where you stare at two identical pieces of test code, identical except that one issues "inferred existential type" at the first line.

      val ref = new SimpleSpec with Reference with Property {
        type ThisCommandLine = SpecCommandLine
        protected def creator(args: List[String]): ThisCommandLine = new SpecCommandLine(args) 
        //def referenceSpec: Reference = this    // perniciously, the type is required
        def referenceSpec = this                 // not even used, could be ??? here
        override def propMapper = new PropertyMapper(this)
      }

This could also be interesting to speed up compilation by using the inherited type rather than type checking the RHS. The decision point is in assignTypeToTree.

Too late for 2.13, though.

I'm not sure if I should prefer the behavior of dotty...
If you're defining public API, especially non-final, you should give it an explicit return type.

I've opened a ticket suggesting the Scala 3 migration guide mention this: scalacenter/scala-3-migration-guide#181

closing as out of scope for Scala 2

Worst possible resolution. Did I ever PR this? This is so obvious etc.

perhaps someone would like to PR it under -Xsource:3?

My duplicate ticket from 2013 was #8159 in case that inspires someone. I thought this already had a PR by lrytz or adriaanm.

I came across this again because test https://github.com/scala/scala/blob/2.13.x/test/files/neg/t2441.scala for #2441 is fixed in Scala 3 under this change. I had forgotten I'd just been here during pandemic.

linked PR changes the behavior, but only under -Xsource:3