Allow calls to init without leading self. or super.

[ahoppen]

ASTGen will need to diagnose the missing self. or super.

[ahoppen]

@swift-ci Please test

[DougGregor]

I'm surprised this is diagnosed in the syntax tree at all. Why doesn't this belong in semantic analysis?

[ahoppen]

IIUC the root cause is that we always consider init as the start of a declaration. If you want to call init on self, you need to disambiguate by prefixing it with self.. At least that’s how the old parser operated and I wanted to maintain this behavior.

Given that init is only valid as a type member and not in code blocks, we could change the rules a little bit and only consider init as the start of a declaration if we are parsing a MemberDeclList. This would make the following a valid parse (it currently isn’t), which will need to be diagnosed in ASTGen. WDYT?

class Foo {
  init() {}
  init(param: Int) {
    init()
  }
}

CC @CodaFi

[DougGregor]

Given that init is only valid as a type member and not in code blocks, we could change the rules a little bit and only consider init as the start of a declaration if we are parsing a MemberDeclList. This would make the following a valid parse (it currently isn’t), which will need to be diagnosed in ASTGen.

I'd be okay with this. It's not context-sensitive in any real sense, and it would be better for recovery to have init() be an expression wherever we allow them.

[ahoppen]

Updated the PR to allow calls to init without self. or super.

[ahoppen]

@swift-ci Please test