codegen: can we implement ascii tree

[tinovyatkin]

Something like https://github.com/Robert42/pest_ascii_tree into our codegen feature?

[tinovyatkin]

Investigated against the linked crate, ANTLR 4.13.2, and current main (bb9a3ca75). Yes, this is doable for ANTLR parser/combined grammars; there is no theoretical blocker, so I am leaving the issue open.

The linked pest_ascii_tree is not doing code generation or grammar analysis. Its core recursively walks Pest's completed Pairs, obtains each rule name, and renders matched text at leaves.

The equivalent grammar-independent data already exists here:

• Node exposes node kind and ordered children.
• RuleNodeView exposes rule index, alternative number, token bounds, and children; terminal and error nodes expose token text.
• Generated metadata already supplies rule names, and the runtime already renders the same tree in ANTLR's LISP form via to_string_tree_with_names.
• The Kotlin/JavaScript/TypeScript parity dumpers already prove that an arbitrary generated grammar can be rendered by walking only this generic surface; for example, the TypeScript dumper.
• Upstream ANTLR's own Trees.toStringTree is likewise just a grammar-independent tree traversal. A branch-style ASCII layout is another presentation of the same model.

It would not be byte-for-byte identical to Pest semantics without explicit formatting choices. ANTLR parse trees contain parser-rule contexts plus terminal/error nodes; lexer rule names are not tree nodes, hidden/skipped tokens are not children, recovery can add error nodes, and transformed left-recursive rules display the tree ANTLR actually constructs. A lexer-only grammar has a token stream but no parser tree. These are representation boundaries to document, not blockers.

So this is viable for every retained ANTLR parser tree. Conceptually it is a generic runtime formatter; codegen already provides the grammar metadata needed to label it.