Comma in `extends` or `derives` clause treated as parameter separator in indentation blocks

[s5bug]

Compiler version

3.3.5
3.6.3
3.6.4-RC1

Minimized code

identity(
  identity:
    class X extends AnyRef, Serializable
    ???
)

Output

',' or ')' expected, but identifier found

Expectation

With braces, this code compiles:

identity(
  identity {
    class X extends AnyRef, Serializable
    ???
  }
)

[som-snytt]

The useful outdent supports (from the unit test)

  assert(
    if x > 0 then
      true
    else
      false, "fail")

The syntax does not distinguish "term commas" from "type commas".

Similar

  assert(
    identity:
      true, "ok"
  )

[som-snytt]

The provisional proposal is to require the comma at EOL if the indent has a colon prefix. So the example would have to be written:

  assert(
    identity:
      true,
    "ok"
  )

That preserves the comma and the spirit of indents in parens.

Certain other tokens work similarly:

  identity(
    x =
      class X extends AnyRef, Serializable // ok, not at EOL and follows =
      27
  )

[som-snytt]

I marked the PR ready for review, to mean discussion.

It is dialed back to only consider colon (as prefix of indented region) for the "comma at EOL" tweak.

The PR shows examples of misleading indentation after => which remain acceptable.

Added a test to show that "arrow eol" is not equivalent to "colon eol" with parameter on next line:

  def test =
    f(
      g: x =>
        x + 1, 42 // ok f(g(_ + 1), 42)
    )

as opposed to

    f(g:
        x =>
          x + 1, 42 // error, arg to f must be on next line
    )

but it is agnostic about the indentation of the arg to f:

    f(g:
        x =>
          x + 1,
          42
    )

which is the same as

    f(g(_ + 1),
          42
    )

The arg to k must be indented wrt g not k (3.6.3):

    f(
      g(_ + 1),
k:
        println(s"calculating...")
        42
    )

[Sporarum]

identity(
  identity:
    class X extends AnyRef, Serializable
    ???
)

To me this makes a lot of sense, I think we should have forbidden the collon syntax for multi-parameter clauses anyways