Fix NoAccessLevelOnExtensionDeclaration to update members inside #if blocks.

[allevato]

Fixes #966.

This rewrites the rule entirely to take advantage of the natural recursion of visiting the extension's children instead of trying to rewrite and replace the member list at the extension level.

I've tended to avoid the idea of writing stateful format rules like this, because pushing state down from a node to its children would normally require maintaining a stack of information. But in this case, the problem is a lot simpler because extensions can't be nested, so there's never more than one level of state to track.

[ahoppen]

I’m sorry to disappoint you but nesting extensions is not a parser error. The following parsers fine

extension A {
  extension B {   
  }
}

Nested extensions only get diagnosed at type-checking time, so I think we do need to handle them…

[allevato]

I’m sorry to disappoint you but nesting extensions is not a parser error. The following parsers fine

extension A {
  extension B {   
  }
}

Nested extensions only get diagnosed at type-checking time, so I think we do need to handle them…

Ugh, yeah, you're right.

Do you have a preference for how you'd like them handled? We could manage the stack and do the same thing for nested extensions that we would for top-level ones, or we could just ignore nested extensions because they're obviously broken.

[ahoppen]

I would just not walk into nested extensions. Leaving them as-is is probably the best thing we can do.

[allevato]

I would just not walk into nested extensions. Leaving them as-is is probably the best thing we can do.

Updated. Instead of using a simple Optional for the state, I created my own State enum. It's technically isomorphic to an Optional (I thought I would need a third state for internal, but realized not because I would be skipping children in that case), but I think it still helps to clarify what's happening.

[ahoppen]

Looks good to me. One suggestion but it’s not blocking.

[ahoppen]

Instead of overriding all these functions, would it make sense to add a visitAny and call applyingAccessModifierIfNone from there?

[allevato]

I just gave that a go and I think it ended up harder to follow. visitAny would have to do a check like "if the node isn't a decl, or if it's an ExtensionDecl or IfConfigDecl, return nil to allow the regular traversal", and it would also have to first call the super implementation because that's where we check if there's an applicable swift-format-ignore directive.

I think it's cleaner to just explicitly list out the decls we care about, even though it means having them listed in two places.

[ahoppen]

OK, I thought that visitAny would have just checked if the node conforms to WithModifierSyntax. But maybe it’s not as easy as that.