Allow anonymous implicit values

[puffnfresh]

Provide support for the following syntax for introducing a new implicit val without assigning a name to it.

implicit val _ = ...

and

implicit val _: TypeClass[Foo] = ...

Incidentally, I just realized that you can currently write:

implicit val _ = foo

as long as you're not at the top-level in the REPL, and you've not already bound something to _. As desired, you've got no way of referring to it by name afterwards, but it's not a solution because you only get one implicit per scope.

[propensive]

My addendum about there being no way to refer to it subsequently is false. You can always refer to using backticks, e.g.

QualifiedScope.`_`

[paulp]

The "clever" trick of actually having a name called _ in the namespace has already shown its limits. Give the poor users a gensym.

[propensive]

I'm in no way advocating it.

[puffnfresh]

It's a good point though, if we're preserving backwards compat, then we need to continue exposing _ as a val - right?

[paulp]

Classic.

object o {
  val _ = 1
  val b = 2
}
object p extends App {
  { import o.`_` ; println(b)   } // ok (!)
  { import o.`_` ; println(`_`) } // ok
  { import o._   ; println(b)   } // ok
  { import o._   ; println(`_`) } // error: not found: value _
}

[puffnfresh]

@paulp wow, what?

[propensive]

I don't even want to contemplate how ALL of those examples are true.

[paulp]

I'll give you a hint:

final val WILDCARD: NameType = "_"

% ack --nogroup --no-filename --no-color  'name\.toString' src/ | wc -l
      85

85 occurrences of the literal character sequence "name.toString". Just imagine if sometimes the name isn't in a variable called "name", or if it's "" + name or s"$name". No that's crazy, probably just the 85.

[gigamonkey]

Wow. Are there legitimate uses where you peer through the anonymity of _? I always thought/hoped it was truly totally anonymous, i.e. caused a fresh, unnamable symbol to be used or some such. (Is that what you meant by "give the poor users a gensym", paulp?)

[paulp]

@gigamonkey no, it is essentially the opposite. _ is much like any other name. No, there are no legitimate uses for it, nor are there any intentional uses for it. It's just an incredibly leaky "abstraction", not that it even rises to the level of abstraction.

However it's still like that because (here we go again) I tried to fix it and that was reverted, because I forgot about java. https://issues.scala-lang.org/browse/SI-6426

Yes, that's what I meant about gensym.

[paulp]

See also https://issues.scala-lang.org/browse/SI-7691 for more fun with underscores.

[gigamonkey]

Hmmm. It seems like Scala could compile each _ used as a pseudo-identifier as a gensym but still allow _ (bah, markdown: that's an underscore in backticks) to refer to the variable named "_" for Java interop. Or am I missing something? (Totally likely.)

[paulp]

@gigamonkey No, you're not missing anything. That's what it should do. With scala the difficulty is always concentrated around making the implementation do what you want.

[som-snytt]

This is tangential, but I was going to mention that at least this is no longer true:

$ scala210
Welcome to Scala version 2.10.4 (OpenJDK 64-Bit Server VM, Java 1.7.0_65).
Type in expressions to have them evaluated.
Type :help for more information.

scala> (1 to 9).toList match { case List(`_`*) => true }
res0: Boolean = true

But instead it's become:

$ scala
Welcome to Scala version 2.11.2 (Java HotSpot(TM) 64-Bit Server VM, Java 1.8.0_11).
Type in expressions to have them evaluated.
Type :help for more information.

scala> (1 to 9).toList match { case List(`_`*) => true }
<console>:8: error: unsupported pattern: Star / p3: List[Int] @ (`_`)* (this is a scalac bug.)
              (1 to 9).toList match { case List(`_`*) => true }
                                                   ^

So I have to revisit a previous conjecture and propose unsupported pattern as the new nosymbol has no owner.

[propensive]

p3? That's a new one for me...

[milessabin]

I think exposing something gensym-like is probably the safest option ... though can we call it freshName please? (I lean towards ML rather than Lisp ;-). That's also what it's called in the reflection API.

[aryairani]

Is there a Java program that uses _ as a "perfectly valid identifier" anyway? :-\

[som-snytt]

Someone used _ as an enum element. I noticed because compiling with Java 8.

[milessabin]

To resurrect this issue, please rework it as an issue/PR against Lightbend Scala (ie. scala/scala).