SyntaxNodes: DotSelfExpr should have optional base expression.

[nkcsgexi]

rdar://46935325

[nkcsgexi]

@swift-ci please smoke test

[nkcsgexi]

@swift-ci please smoke test and merge

[slavapestov]

This is a keypath, not a DotSelfExpr. I don't think its correct that DotSelfExpr's base is optional; you should instead parse keypaths differently. CC @jckarter

[nkcsgexi]

We parse them as DotSelfExpr too for AST, see here.

[slavapestov]

I see it parses like this:

        (keypath_expr type='<null>'
          <<null>>
          (dot_self_expr type='<null>'
            (key_path_dot_expr implicit type='<null>')))))))

So the base is a KeyPathDotExpr. Do you not have an equivalent in SwiftSyntax? It would be better if the base was never optional here and with MemberExpr also.

[nkcsgexi]

Syntax tree doesn't have equivalent implicit nodes in general. All nodes should cover at least one underlying token.

[nkcsgexi]

The only exception is collection syntax like CodeBlockItemList, which can be an empty set of items.

[akyrtzi]

a.self is creating a DotSelfExpr syntax node while .self is creating a ImplicitMemberExpr one. If we want to be consistent here then shouldn't \.self be a KeyPathExpr wrapping an ImplicitMemberExpr ?

[akyrtzi]

Also there is a difference in .a creating a ImplicitMemberExpr node while \.a is a MemberAccessExpr wrapped by KeyPathExpr.

[akyrtzi]

a.self is creating a DotSelfExpr syntax node while .self is creating a ImplicitMemberExpr one. If we want to be consistent here then shouldn't \.self be a KeyPathExpr wrapping an ImplicitMemberExpr ?

After further discussion with Xi, this is explained by the parser treating these differently, .self being a unresolved_member_expr while \.self creating a dot_self_expr.

[slavapestov]

Ok, makes sense. Since SwiftSyntax is trying to be more uniform and elegant perhaps it would make sense if .self in all contexts was a MemberRefExpr with an identifier of self? Does it make sense for it to be its own thing?

[nkcsgexi]

yeah, i think this suggestion makes sense. We could reuse existing MemberAccessExpr to cover DotSelfExpr. We could also use MemberAccessExpr to represent ImplicitMemberExpr to further reduce the type complexity.

[nkcsgexi]

@swift-ci please smoke test