The underlying parser is just tree-sitter, which is a reusable (and excellent) parsing library.

Difftastic then converts the tree-sitter parse tree to a simpler s-expression style format (see https://difftastic.wilfred.me.uk/parsing.html#simplified-syn...), and computes differences on that.

I'm just trying to clarify that I'm not generating conventional 'unified diff' patches, so I can provide a nicer interface (e.g. line numbers).