Add `mdast_util_to_hast` subcrate

[pd4d10]

Would it be supported in the future plan?

[wooorm]

Feel free to take the code for that now from here: https://github.com/wooorm/mdxjs-rs/blob/main/src/mdast_util_to_hast.rs.

I don’t really know how to best expose those parts. And you will likely need more parts (hast-util-to-html). I don’t think they make a lot of sense here. Depends on how people want to use these things!

[kwonoj]

Just fyi, I'm also in need of similar api exposure from mdxjs-rs, though I'm also not sure final shape of surface yet (under experimenting now).

@wooorm Is this something acceptable in upstream mdxjs-rs PR to expose few interfaces as public? or the recommendation is copy-code-as needed?

[wooorm]

Can you tell me more about what you are working on?

My previous comment still stands: you will likely need more things. I’d probably want to expose several unist, mdast, and hast utilities, each in a different monorepo for their AST?

Keeping it hidden for now means that some of the AST can still be improved. I’d particularly first like to:

• Use traits on each node instead of specifying things globally (e.g. for context see this), that would allow unist utilities (such as this) to work on both mdast and hast
• Make AST nodes serializable in a good way: https://github.com/wooorm/markdown-rs/pull/30

After those two, I think it should be a stable API that can be exposed either from the crates currently or later from separate packages

[kwonoj]

Sure, the current work I'm trying to do is integrate mdx support in turbopack (https://github.com/vercel/turbo).

I'm still in the process of figuring out what's necessary piece, so I may wrong, but so far I came to conclude it requires 2-stage process for parse(input, option) -> AST / transform(AST, option) as separate. This requires some internal access to existing mdxjs-rs to split things up and create separate api surfaces.

Again, please note this is not a final form and reason I mentioned though I'm also not sure final shape of surface yet (under experimenting now).. I would like to get more clarity before making any actual upstream effort.

[wooorm]

Cool, interesting!

so far I came to conclude it requires 2-stage

You mean that turbopack enforces this, right? So it has several hooks for plugins (likely also serialize/stringify?).

• Do you have a link to some docs on these, or some type definitions perhaps?
• the big question for these signatures is: what does transform want to do? Because for MDX, it could mean: a) transform markdown (mdast/remark), b) transform html (hast/rehype), c) transform javascript (estree/esast/recma, or maybe turbopack enforces swc?)

[kwonoj]

Do you have a link to some docs on these, or some type definitions perhaps

afaik no clear docs around this unfortunately. I'm reading existing implementation to support other assets (js, css) - i.e there is a fn parse definition for ecma support https://github.com/vercel/turbo/blob/main/crates/turbopack-ecmascript/src/parse.rs#L129 returns parsed result, and it is being transforme lazely elsewhere.

the big question for these signatures is: what does transform want to do?

This is great question and I do not have concrete answer yet, I'd like to play with my wip first to see how it looks like.

[wooorm]

For a first MVP, I’d go with just a parse hook:

• Use mdxjs-rs core to turn a string of MDX into a string of JS
• Parse that with SWC again to get a JS AST Then nothing needs to happen in the transform hook, and hopefully you can somehow tell turbopack to serialize the AST with SWC as a string of JS later

For a second MVP, I’d use a fork that specifies a parse hook with: https://github.com/wooorm/mdxjs-rs/blob/d0a15b116d6a09e98149188eb190537a2faeac9e/src/lib.rs#L113-L114 A transform hook with: https://github.com/wooorm/mdxjs-rs/blob/d0a15b116d6a09e98149188eb190537a2faeac9e/src/lib.rs#L115-L122. And a serialize hook with: https://github.com/wooorm/mdxjs-rs/blob/d0a15b116d6a09e98149188eb190537a2faeac9e/src/lib.rs#L124.

Assuming we have MDX plugins soon, you could hopefully access those from options in the transform hook, and apply some of them on the markdown AST, some on the HTML AST, and some at the end in the JS AST.

I’d be open to splitting these steps in 3 functions, and exposing those from mdxjs-rs later if your MVP 2 works?