Note that this common format for source maps isn't useful for us here, because it correlates a textual original source with a textual generated source. We need to correlate positions in the original textual source with elements in the AST.

Something like:

sourcemap = [ [1, 3, 6], [2, 3, 5] ]

to indicate that the node occupies columns 3-6 of line 1 and columns 3-5 of line 2.

I guess, you need this package https://github.com/mozilla/source-map . It's used everywhere to build soursemap data and includes support for incremental (nested) builds with flatten at the end. Don't know about other languages.

Though, that will significantly decrease speed and will require to rewrite regex replacements because those break positions. IMHO, it's better to rewrite parser first to be modular/pluggable.