SignatureHelp gives apply of the current symbol instead of enclosing apply

Compiler version

3.1.1

Explanation

I hit on this with Metals but then testing in Dotty with the SignatureHelp tests it suffers from the same issue, so the fix should probably be here.

Given the following code with the @@ signifying a trigger of the user to get signature help.

Li@@st(1, 2,  3)

The user will actually get the signature help of the apply on the List. That may not seem that problematic, but when you have something like this:

def doThing(myList: List[Int]) = ???

doThing(L@@ist(1, 2, 3))

The user is expecting to see the signature help for doThing, not the apply on List.

Example test

Here is what I used to test this:

  @Test def applyWhenYouDontWantIt: Unit = {
    val signature =
      S("apply[A]", Nil, List(List(P("elems", "A*"))), Some("CC[A]"))
    code"""|object O:
           |  def doThing(myList: List[Int]) = ???
           |  doThing(Li${m1}st(1, 2, 3)""".withSource
      .signatureHelp(m1, List(signature), Some(0), 0)
  }

Expectation

The test above will pass, when you really want it to fail and to be showing you the info for doThing meaning I care about the myList here.

[tgodzik]

I think we should only show the signature help when we are inside parenthesis, which is where it’s most useful. The purpose of it is to be able to write the correct parameters. For just seeing the signature, we can use hover.

Isn’t the issue here that we don’t try to find the most recent enclosing apply, but use the identifier to calculate the signature help? We should make sure that we are actually inside the args.

[prolativ]

Ok, I think I get it now