Value parameter inference for polymorphic lambdas

[smarter]

Release Notes

Value parameter inference for polymorphic lambdas

It is now possible to write a polymorphic lambda without writing down the types of its value parameters as long as they can be inferred from the context, for example:

val f: [T] => T => String = [T] => (x: T) => x.toString

can now be written more concisely as:

val f: [T] => T => String = [T] => x => x.toString

Detailed Description (not part of the release notes)

Just like

val f: Int => Int = x => x

is inferred to

val f: Int => Int = (x: Int) => x

we now accept

val g: [T] => T => T = [S] => x => x

which is inferred to

val g: [T] => T => T = [S] => (x: S) => x

This requires a substitution step which is tricky to do since we're operating with untyped trees at this point. We implement this by generalizing the existing DependentTypeTree mechanism (already used for computing dependent result types of lambdas) to also be usable in any other position inside the lambda.

We rename the tree to InLambdaTypeTree at the same time for clarity.

Note that this mechanism could also probably be used to allow closures with internal value parameter dependencies, but we don't attempt to support this here.