Under -Xsource:3, warn when eta-expanding SAMs

[dwijnand]

Dotty emits the following:

-- Warning: target/tests/EtaX/EtaX.meth1/EtaX.meth1/EtaX.meth1.01.scala:10:24 --
10 |  val t5b: Sam1S      = meth1                   // ok, but warning
   |                        ^^^^^
   |method meth1 is eta-expanded even though Sam1S does not have the @FunctionalInterface annotation.
1 warning found

(given the following source:)

trait Sam1S { def apply(x: Any): Any }

class Test {
  def meth1(x: Any) = ""
  val t5b: Sam1S      = meth1                   // ok, but warning
}

So it makes sense for Scala 2.13 to do the same, under -Xsource:3.

Now it actually also emits in Dotty under -source 3.0-migration (which used to be called -language:Scala2Compat), so there's an argument that it should be on by default in 2.13... But let's do -Xsource:3 at the very least.

[joroKr21]

I've never seen anyone add @FunctionalInterface to Scala traits.

[joroKr21]

But I have seen a lot of code use this feature, perhaps courtesy of Intellij inspection suggesting it.

[joroKr21]

Or maybe I got this wrong - is this about SAM without @FunctionalInterface in general or only about eta-expansion? If we define the function explicitly there will be no warning?

[dwijnand]

Hey, @joroKr21. Yeah, that's right: this is only about eta-expanding methods where a SAM is expected. I don't know the history around @FunctionalInterface but I think a call was made that those will continue to eta-expand cleanly, while others eventually (somewhere in 3.x) won't.

[smarter]

I don't know the history around @FunctionalInterface but I think a call was made that those will continue to eta-expand cleanly, while others eventually (somewhere in 3.x) won't.

I believe we'll keep the current behavior forever, the warning is there to prevent confusion and is easily solvable by making the eta-expansion explicit, see discussion in scala/scala3#4364

[neko-kai]

@smarter @dwijnand Then the error message should mention explicit expansion first as it's totally confusing as to what's the purpose of the warning otherwise.

[dwijnand]

Are you saying the "write out the equivalent function literal" doesn't convey "explicit expansion" and/or that it's too late in the message?

I'd be happy if we were to tweak the messaging. Want to send a PR with what you have in mind would work better?

[neko-kai]

@dwijnand The former, "write out the equivalent function literal" is extremely heavy and doesn't give any hint that appending underscore f _ will fix the issue - by contrast the error messages when automatic eta-expansion cannot be performed due to overloading give a very clear guide - 'add underscope and it will compile'.

Want to send a PR with what you have in mind would work better?

I am strictly against this warning in its entirety, both in Scala 2 and in Dotty, i'll give an example below why. This particular change makes -Xsource:3.0 come with a large amount of un-disableable warnings, which is regrettable if i need to access the changes there such as fixed contravariant typeclass resolution.

I use eta-SAM syntax mostly to concisely implement type classes, e.g.:

trait Encoder[A] {
  def encode(a: A): Json
}
object Encoder {
  implicit val jsonEncoder: Encoder[Json] = identity
}

I don't find the linter complaints very relevant in this case, neither do i find the original example in scala/scala3#4364 (comment) surprising or justifying the warning, so I'm simply opposed to the warning.

[SethTisue]

@dwijnand We're release-noting #9049, so we don't need to also release-note this one... right?

[dwijnand]

Yep, these are back to error again.