Unused identifiers give weird errors when something comes after them

[idillon-sfl]

An unused identifier is invalid code and is expected to give errors. However, when some valid code come after it, the errors get weird and it make it difficult to provide something useful to the user.

This one is fine:

BAR='bar'
FOO

(recipe [0, 0] - [1, 3]
  (variable_assignment [0, 0] - [0, 9]
    (identifier [0, 0] - [0, 3])
    (literal [0, 4] - [0, 9]
      (string [0, 4] - [0, 9]
        (string_content [0, 5] - [0, 8]))))
  (ERROR [1, 0] - [1, 3]
    (identifier [1, 0] - [1, 3])))

These are not:
1.

FOO
BAR='bar'

(recipe [0, 0] - [1, 9]
  (variable_assignment [0, 0] - [1, 9]
    (identifier [0, 0] - [0, 3])
    (ERROR [1, 0] - [1, 3]
      (python_identifier [1, 0] - [1, 3]))
    (literal [1, 4] - [1, 9]
      (string [1, 4] - [1, 9]
        (string_content [1, 5] - [1, 8])))))

FOO
BAR() {
    
}

(recipe [0, 0] - [3, 1]
  (function_definition [0, 0] - [3, 1]
    (identifier [0, 0] - [0, 3])
    (ERROR [1, 0] - [1, 3]
      (python_identifier [1, 0] - [1, 3]))))

FOO
python bar() {

}

(recipe [0, 0] - [3, 1]
  (function_definition [0, 0] - [3, 1]
    (identifier [0, 0] - [0, 3])
    (ERROR [1, 0] - [1, 10]
      (python_identifier [1, 0] - [1, 6])
      (ERROR [1, 7] - [1, 10]))))

In each cases the FOO identifier is wrongly put inside a function_definition. A python_identifier also comes out of nowhere in 1. and 2.

[amaanq]

This is difficult - as much as I'd like to fix this, I think the solution is to either rely on better output from incremental parsing (as the USER types, the tree isn't as malformed when just immediately parsing bad code), or to allow this as an invalid rule and it be denoted as such

[idillon-sfl]

I noticed the following case does not give error, but still gives wrong parsing. It seems related.

foo:
python() {
}

(recipe [0, 0] - [2, 1]
  (function_definition [0, 0] - [2, 1]
    (identifier [0, 0] - [0, 3])
    (override [0, 3] - [1, 6]
      (identifier [1, 0] - [1, 6]))))