Considering these points:

• A couple of additional nodes are added for clarity (i.e., ArgStar, Not, PinnedBegin, etc.).
• Every node has location information attached to it (as opposed to just the scanner event nodes).
• There's a standard interface for descending the tree (child_nodes).

If we're going to add new nodes for clarity, I think we could also take the liberty to add a common ancestor to all nodes. This common ancestor could hold the location and comments. Maybe even define an empty child_nodes method that would be overridden by the sub classes.

Having a common ancestor to all nodes has at least two advantages when using the resulting tree to write tools:

1. If you work with typing (be it RBS or Sorbet), this makes signatures easier to write. Consider this:

   class Visitor
     def visit: (node: Node) -> void
   end

   instead of:

   class Visitor
     def visit: (node: Alias | ARef | ARefField | ArgParen | ...) -> void
   end

   and it lets you properly/easily type the accept method:

   class Node
     def accept: (v: Visitor) -> void
   end

   instead of:

   class Alias
     def accept: (v: Visitor) -> void
   end
   
   class ARef
     def accept: (v: Visitor) -> void
   end
   
   ...

2. When writing tools it allows you to attach behavior to all nodes at once without requiring to monkey patch them or rely on meta-programming:

   class Node
     def to_some_format_i_need
       {
         kind: class.name,
         loc: location.to_s,
         children: child_nodes.map(&:to_some_format_i_need)
       }
     end
   end

   instead of defining this method on each node separately.

Awesome work on this + would be great to have something like this as part of ripper. Couple of thoughts:

• Any reason the node classes aren't in their own files? It's sort of nice to have them live near their handlers, but having an 11k+ line file really reduces the accessibility of this code.
• The lack of test coverage for ripper has been a pain-point for me. Would be great if some of the tests from syntax_tree could come with this, maybe with the goal to add more. I think with a good test suite it's possible this feature could reduce the total maintenance burden of ripper, since breaking changes to Ruby's AST would be caught earlier.