As this module is the first one in Redwood which is ESM only (I ported it from Deno) then it was the first case in the codebase where ESM/CJS interact.

This means it was the first time, from the looks of the rest of the codebase, that import("thing") was used. As we target versions of node which all have support for that, I have dropped the babel backporting for the feature so it can be used natively.

Just to illustrate this change for others, here's some example source code from the CLI:

export const handler = async (options) => {
  const { handler } = await import('./buildHandler')
  return handler(options)
}

And here's the output dist with our current Babel config (Babel transforms await import to an async-like require):

const handler = async options => {
  const {
    handler
  } = await _promise.default.resolve().then(() => (0, _interopRequireWildcard2.default)(require('./buildHandler')));
  return handler(options);
};

And here's the output dist with this change (the code stays the same; Babel changes the spacing a bit but that doesn't matter):

const handler = async options => {
  const {
    handler
  } = await import('./buildHandler');
  return handler(options);
};

All of the work happens in the sdl-codegen side, there's a good argument that this is flawed (because what if Redwood changes the shape of this paths object) but I copied the types over into the lib.

I think it's reasonable that at some point I chop this down to just the types from Path which are used, and then you aren't accidentally constrained on the shape of the object for anything but the API side.

You use the dmmf to grab this info, but it doesn't include the comments! So, I can't re-use that approach.

Without this conditional, Jest wants us to pass --experimental-vm-modules to node. I tried that and most of our test suites actually worked save for the CLI and codemods. But I figured it may be easier to just disable this config for testing since we only really have one ES module we're importing in the entire framework