Can ToString[.] type be made applicable to arbitrary type by converting its symbol name as singleton?

[tribbloid]

I have posted a stackoverflow question on this:

https://stackoverflow.com/questions/66837880/how-to-get-the-name-of-a-class-as-a-string-literal-at-compile-time-using-shapele

According to Prof. @DmytroMitin, it appears that a whitebox macro is the only option, fortunately it is not too complex, in dotty it may be introduced as a compiletime ops later

This feature, combining with RequireMsg[.] type, can be used to throw a customised ops error at compile-time, using a string interpolated with custom user type (https://twitter.com/soronpo/status/1275014596034809856). Similar to how @implicitNotFound error message is interpolated in native scala:

@implicitNotFound(msg = "Cannot prove that ${From} =:= ${To}.")
sealed abstract class =:=[From, To] extends (From <:< To) with Serializable {
...
}

Do you think this will be useful? I can submit a PR shortly but I may need some guidance to circumvent the NonLiteral constraint

[soronpo]

It shouldn't be ToString, but GetNameOfSym. I think it's ok to add. But maybe another way is better- the way string interpolation is currently done inside the implicitNotFound message. We just add this ability directly to the RequireMsg functionality.

[tribbloid]

Thanks a lot, I didn't find any test case for RequireMsg with such feature, is there a class name I should focus on?

[tribbloid]

Oh sorry I misunderstood it, I thought it is already being done. We still need to write it.

So you prefers to allow RequireMsg to take more type arguments and expand each ${} in the second arg (Msg) with symbol name of 3+ args? How many args can you support at maximum (without using annotation like @ImplicitNotFound)?

The GetNameOfSym seems to be a bit easier to me.

[soronpo]

No. What I mean is that message string inside it will support "${sym}"

[tribbloid]

I see, but how does RequireMsg know what "${From}" or "${To}" refers to? They are just type arguments that are not in the scope

[soronpo]

The same way RequireMsg is determines what symbol to apply the implicitNotFound annotation on.
Actually, I haven't tried it, but I wouldn't be surprised if it works as-is.

[soronpo]

Yes, apparently it works:

trait Foo[T]
object Foo {
  implicit def ev[T](implicit r : RequireMsg[false, "Bad Type: ${T}"]) : Foo[T] = ???
}
implicitly[Foo[Int]] //error: Bad Type: Int

[tribbloid]

Nice! So it addressed the most simple case.

For more complex case it we may still need the GetNameOfSym. Invoking type members by A#B doesn't work though:

  trait HasM {

    type M = Int
  }

  trait Foo[T]

  object Foo {

    implicit def ev[T <: HasM](
        implicit
        r: RequireMsg[false, "Bad Type: ${T#M}"]
    ): Foo[T] = ???
  }
  implicitly[Foo[HasM]] //error: Bad Type:

It can get even more hairy if ${T} is replaced with all kinds of dependent types. So it make sense to be more specific about the exact operation to string

[soronpo]

I think it's maybe better to add the dependent type interpolation support in the compiler itself.
I propose opening a thread on contributors.scala-lang.org to discuss this further.

[tribbloid]

@soronpo I'm not a compiler's guy. But to me it appears to be never documented as a scala 2 feature or language specification. Even the original @ImplicitNotFound argument doesn't list it as a feature, I could only speculate that from the example in =:=. You may have a better chance to convince dotty team to support it.

NameOfSym or NameOfClass on the other hand can be used to ensure that a type must be from a certain package or so. Maybe worths a try.

Also it could serve as the default ToString[.] impl for custom operands defined by OpIntercept (https://twitter.com/soronpo/status/1275014596034809856/photo/2)