Invalid generic signature for type params bounded by primitive

[scf37]

Compiler version

3.1.6

Minimized code

def foo[A <: Int]: A = ???

Output

public <A> A foo() {
    throw Predef$.MODULE$.$qmark$qmark$qmark();
}

Expectation

public int foo() {
    throw Predef$.MODULE$.$qmark$qmark$qmark();
}

It is important for performance of code dealing with types carrying meta-information like

type Validated[A] = A
opaque type Max[A, max] <: A with Validated[A] = A

[nicolasstucki]

That example seems to compile using the primitive type. Though a slightly different example does not.

class Foo:
  def foo[A <: Int]: A = ???
  def bar[A <: Int](a: A): A = ???

public class Foo {
  public Foo();
  public int foo();
  public <A> A bar(A);
}

[smarter]

Both of the methods here haveA erased to Int (see the descriptor in the output of javap), the problem is that the generic java signature we generate for bar is incorrect (this is what javap uses to display the method signature but it's not the actual bytecode signature, it's only used by javac), A shouldn't be preserved in Signature when its upper-bound is a primitive type:

% javap -p -v Foo.class
[...]
  public int foo();
    descriptor: ()I
[...]
  public <A extends java.lang.Object> A bar(A);
    descriptor: (I)I
[...]
    Signature: #28                          // <A:Ljava/lang/Object;>(TA;)TA;
[...}

The same bug exists in Scala 2: scala/bug#9846

For anyone interested in working on this, the generic signature logic is in https://github.com/lampepfl/dotty/blob/main/compiler/src/dotty/tools/dotc/transform/GenericSignatures.scala

[scala-center-bot]

This issue was picked for the Issue Spree No. 24 of 29 November 2022 which takes place in a week from now. @jan-pieter, @nmcb, @TheElectronWill will be working on it. If you have any insight into the issue or guidance on how to fix it, please leave it here.