Unary function with argument of type `Unit` is not type-safe

[otto-dev]

reproduction steps

using Scala 2.13.6,

scala> val f = (a: Unit) => println(s"Received $a")
val f: Unit => Unit = $Lambda$2279/411876574@4ee6291f

scala> f("foo", "bar", "baz")
Received ()

problem

The compiler should not allow the arguments ("foo", "bar", "baz") to be passed to a function that expects one Unit argument. This is particularly problematic for higher-order functions that may return a unary A => Unit function, with A being Unit in situations where the first function argument is irrelevant.

[som-snytt]

scala> f(42, 17) // print
                                                                f.apply({
  scala.Tuple2.apply[Int, Int](42, 17);
  ()
}) // : Unit

[eed3si9n]

As annoying and un-typesafe as it is, this is perfectly legal under the current specification. SLS 6.26.1 Value Conversions

Value Discarding

If 𝑒 has some value type and the expected type is Unit, 𝑒 is converted to the expected type by embedding it in the term { 𝑒; () }.

[dwijnand]

True, but this is under auto-tupling. I'm sure it's spec-legal to make this not compile, right?

[som-snytt]

➜  ~ scala -Werror -Wvalue-discard -Xlint
Welcome to Scala 2.13.7 (OpenJDK 64-Bit Server VM, Java 17).
Type in expressions for evaluation. Or try :help.

scala> val f = (u: Unit) => println(s"[$u]")
val f: Unit => Unit = $Lambda$1122/0x0000000801086220@203d1d93

scala> f(42, 17)
        ^
       warning: discarded non-Unit value
        ^
       warning: adapted the argument list to the expected 2-tuple: add additional parens instead
               signature: Function1.apply(v1: T1): R
         given arguments: 42, 17
        after adaptation: Function1((42, 17): Unit)
error: No warnings can be incurred under -Werror.

[SethTisue]

I had expected to label this "fixed in Scala 3", but I'm puzzled that Scala 3.1.0 doesn't complain, even under -source:future. I thought we'd agreed that adapted arguments were being phased out the language. Would someone like to dig around in the Scala 3 issue tracker and see if a bug ought to be opened?

Closing this ticket, since it's working as designed, even though we ended up regretting the design. fwiw, I'd never write a line of Scala 2 without -Xlint, not even in the REPL.

[som-snytt]

@SethTisue I also tried it twice. The spec under autotupling links to the suspended PR 4311; see also https://docs.scala-lang.org/scala3/reference/overview.html#dropped-constructs Part of that pandemic saga was linting its sister wart, multi-arg infix. I get my zigs mixed up with my zags.

For this issue, perhaps a narrow lint is called for. Discarding args seems especially pernicious.

[SethTisue]

Okay, shall we reopen it as a "help wanted" enhancement ticket?

(I edited your comment slightly.)

[som-snytt]

Thanks, I get my jigs mixed up with my jags. Oh, I had intentionally not linked the PR to avoid extra noise. Anyway, feel free to reopen. Perhaps someone will step up.

BTW the other nice link there is their refusal to consider the request for "autotuple Tuple1" feature.