Delete SyntaxTransformVisitor

[ahoppen]

Since ASTGen is no longer using it, we can delete it.

[ahoppen]

@swift-ci Please test

[bnbarham]

Will need to remove from CMake as well

[allevato]

Heh, I was experimenting with using this in a swift-format branch to rewrite parts of the pretty-printer 😭 What I'm in need of is a visitor where each visit method is truly recursive instead of splitting visit and visitPost into separate operations.

I could coax SyntaxRewriter to do what I want by just always returning the input node; what would the overhead of that be? If the visit methods always return the same node, would there be any excessive overhead there?

[ahoppen]

@swift-ci Please test

[ahoppen]

Heh, I was experimenting with using this in a swift-format branch to rewrite parts of the pretty-printer 😭 What I'm in need of is a visitor where each visit method is truly recursive instead of splitting visit and visitPost into separate operations.

Why doesn’t SyntaxVisitor work for you? You could just ignore visitPost, right? I feel like I’m missing something.

I could coax SyntaxRewriter to do what I want by just always returning the input node; what would the overhead of that be? If the visit methods always return the same node, would there be any excessive overhead there?

The overhead shouldn’t be big. SyntaxRewriter has a special check to see if the syntax node has changed and if no node is changed, it doesn’t actually do any rewriting.

https://github.com/apple/swift-syntax/blob/1b7b2dbbcf03905ae410f29d54846f45c44c0584/Sources/SwiftSyntax/generated/SyntaxRewriter.swift#L3826

[allevato]

Why doesn’t SyntaxVisitor work for you? You could just ignore visitPost, right? I feel like I’m missing something.

One of the challenges of today's implementation is that because visitation of children is handled by returning .visitChildren/.skipChildren, we're not doing any direct recursion within the visit methods. That makes it error-prone because we can't use recursion and scoping to require that certain formatting tokens are correctly positioned before and after syntax nodes/tokens. For example, every open grouping command should have a corresponding close, and they need to be nested correctly.

Here's the structure I'd like to have instead, using true recursion and closure-based helper methods to define scopes that ensure everything lines up: https://github.com/allevato/swift-format/blob/new-pretty-printer/Sources/SwiftFormat/PrettyPrint/NewTokenStreamCreator.swift#L340-L371

But maybe I'm overthinking this. I guess I could stick with SyntaxVisitor but manually call visit on a node's children and always return .skipChildren from the visit methods to short-circuit the built-in traversal? Do you see any problems with that approach?

[ahoppen]

Oh, I see what you mean now. Using SyntaxRewriter and always returning the node itself should work for you and AFAICT there shouldn’t be any real performance overhead associated with it.

[ahoppen]

@swift-ci Please test