Check out magic-string as a potential tool for generating source maps iteratively https://www.npmjs.com/package/magic-string

I'm having some silly idea of generating an inline source map after the compiler transformation and adding that to the final code that's going to be executed by the vm(mostly).

I currently don't know whether the inline source map will be considered when an error is thrown

It seems the major difference between source code and compiled code is some additional code that wraps it. Hence we should be able to map the lines I guess.
eg. source

fn(state=> state)
fn(state=> {
  state.data.smth = "smth"
  return state;
})

eg. compiled

import {fn} from "@openfn/common"

export default [
fn(state=> state),
fn(state=> {
  state.data.smth = "smth"
  return state;
})
]

mostly the entire source code just sits in the array.

Just some random thought.

Hi @doc-han - I'm actually half way through this story. We get a source mapping solution out-of-the-box with recast - it's just about applying it in the right places in the code.

I've got a prioritised work list for you with plenty of interesting compiler and runtime stuff, and we've got a couple of bigger projects planned too. Hang tight and we'll keep you busy!