Regression in typer for closure body treated as named tuple

[WojciechMazur]

Based on OpenCB failure in scala-native/scala-native - build logs

Related #21767

Compiler version

Last good release: 3.6.0-RC1-bin-20241003-a672e05-NIGHTLY
First bad release: 3.6.0-RC1-bin-20241005-6fa81cf-NIGHTLY

Bisect points to 87cdbc8

Minimized code

@main def Test = {
  var cached: Option[scala.Double] = None
  val spliterator: java.util.Spliterator.OfDouble = ???
  spliterator.tryAdvance((e: Double) => (cached = Some(e)))
}

Output

[error] ./test.scala:4:3
[error] None of the overloaded alternatives of method tryAdvance in trait OfDouble with types
[error]  (x$0: java.util.function.Consumer[? >: Double]): Boolean
[error]  (x$0: java.util.function.DoubleConsumer): Boolean
[error] match arguments (Double² => (cached : Some[Double²]))
[error] 
[error] where:    Double  is a class in package java.lang
[error]           Double² is a class in package scala
[error]   spliterator.tryAdvance((e: Double) => (cached = Some(e)))

Expectation

Should compile

[Gedochao]

cc @odersky

[odersky]

Yes, that will no longer work. Workarounds:

spliterator.tryAdvance((e: Double) => {cached = Some(e)})

Or, better:

spliterator.tryAdvance { (e: Double) => cached = Some(e) }

Or:

spliterator.tryAdvance: (e: Double) => 
  cached = Some(e)

Each of these is clearer than the original.

One thing we can do is a migration hint with a rewrite rule.