newline causes erroneous class parsing

[brandonspark]

Current behavior:

Suppose that I had a class that looked like this:

public class Foo constructor () { 
   < stuff in here >
}

This is meant to be parsed in the grammar as a class declaration, with a primary_constructor. Indeed, we see it parsed as such with the current grammar:

(source_file [0, 0] - [1, 0]
  (class_declaration [0, 0] - [0, 35]
    (modifiers [0, 0] - [0, 6]
      (visibility_modifier [0, 0] - [0, 6]))
    (type_identifier [0, 13] - [0, 16])
    (primary_constructor [0, 17] - [0, 31])
    (class_body [0, 32] - [0, 35])))

If I add a newline, however:

public class Foo
constructor () {
  < stuff in here>
}

it parses as something different.

(source_file [0, 0] - [2, 0]
  (class_declaration [0, 0] - [0, 16]
    (modifiers [0, 0] - [0, 6]
      (visibility_modifier [0, 0] - [0, 6]))
    (type_identifier [0, 13] - [0, 16]))
  (call_expression [1, 0] - [1, 18]
    (call_expression [1, 0] - [1, 14]
      (simple_identifier [1, 0] - [1, 11])
      (call_suffix [1, 12] - [1, 14]
        (value_arguments [1, 12] - [1, 14])))
    (call_suffix [1, 15] - [1, 18]
      (annotated_lambda [1, 15] - [1, 18]
        (lambda_literal [1, 15] - [1, 18])))))

The reason for this ends up being that the newline induces an opportunity for an automatic semicolon, so the token stream looks like:

PUBLIC CLASS "Foo" SEMICOLON CONSTRUCTOR ( ) ...

This makes the parser think that, since a class_declaration can have many optional things, that the public class Foo is a standalone class (with not very many things in it), and the thing that follows it is a call expression of constructor on a lambda or something, I think.

Expected behavior:
The above example should correctly parse to a single class declaration, with a primary constructor, as opposed to two top-level entities. This could be done by suppressing the insertion of the automatic semicolon in such a spot (which may require refactoring and heavier state being carried in the external scanner), or some grammar-level hacking.

Let me know what you think! I am trying to add better support for Kotlin in an open-source static analysis tool, Semgrep (https://github.com/returntocorp/semgrep), and this is blocking my ability to do so.