WIP - Try and transform jsx through compilation

[nojaf]

Super rough experiment about type-checking the new Jsx untyped nodes.
Current attempt is to compile let x = <div className="yow"></div> and print jsx in js_dump.

Current result:

[nojaf]

Hello @cristianoc, curious to hear your thoughts on this!
//cc @zth

[nojaf]

This idea here is to create a record of JsxDOM.domProps and type-check that.
It is of course laser focused to the <div> example.

(children could later be checked separately, as I believe we need to check if those Jsx.element)

[nojaf]

jsx_dom_type_expected isn't quite right here. Could we maybe try and find it in the env?

[nojaf]

We transform this node a couple of times later on, in all the phases but overall the important check happened here.

[cristianoc]

After looking at how complexity builds up, perhaps this was not the best suggestion.
If the result of the normal transformation is a form of application, then that application seems to contain almost all the information needed. E.g. ReactDOM.jsx("div", ...).
Perhaps a simpler way is to add some info to the application, so that it's treated differently after type checking.
That seems easy for things like div, not sure about the other cases.
Ideally, we'd want not to complicate type checking, but at the same time keep enough information to emit code in preserve mode.
Perhaps one can add a bit more information to the application ast node, and if useful modify a bit the ppx too if reconstructing the info for preserve mode is more complex than it needs to be.

Thoughts?

[nojaf]

I agree that this touches on many layers, and the impact of breaking something is considerable.

Marking the transformed application call with additional data could also be effective. I would support adding the original jsx_element node as optional data in Pexp_apply. This information needs to be carried over every time the application gets transformed, as I don't see any way around that.

We could ultimately detect it in Js_dump and reformulate it as JSX. This might be feasible, and I'll experiment with this idea further, see #7387