Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

What is the MDXProvider? #197

Closed
brandonweiss opened this issue Jul 25, 2018 · 6 comments
Closed

What is the MDXProvider? #197

brandonweiss opened this issue Jul 25, 2018 · 6 comments

Comments

@brandonweiss
Copy link
Contributor

The section in the README on the MDXProvider is very short and seems to assume a lot… I don't know what this is, why it exists, what it’s for, etc. Could someone expand on it?

@silvenon
Copy link
Contributor

Does this clear things up?

import { MDXProvider } from '@mdx-js/tag'

import One from './one.mdx'
import Two from './two.mdx'
import Three from './three.mdx'

import { Title, Subtitle } from './body'

const components = {
  h1: Title,
  h2: Subtitle,
}

const AppWithoutProvider = () => (
  <div>
    <One components={components} />
    <Two components={components} />
    <Three components={components} />
  </div>
)

const AppWithProvider = () => (
  <MDXProvider components={components}>
    <One />
    <Two />
    <Three />
  </MDXProvider>
)

@brandonweiss
Copy link
Contributor Author

@silvenon Thanks for helping!

🤔 I think this is heading in the right direction, although there are still some things I don’t understand…

What is the components prop? Is that a React thing? An MDX thing? Does something magical happen with it where it appends components to the end of the component it’s passed into? That doesn’t seem right… but it’s hard to tell because I can't see how components is being consumed inside the child components?

Oh, is the provider a way of passing the same props to every child component? That seems… dangerous? Couldn't you obliterate some props without realizing it? Does it only work with direct children or all children (children of children)?

If that’s what it does… why is it an MDX Provider? It seems like just a generic tool for passing props to all children? There must be something about it I'm missing that ties it into MDX specifically?

@silvenon
Copy link
Contributor

silvenon commented Jul 25, 2018

It is a bit magical, yes, but deliberately, for the sake of simplicity of usage.

Let's use this MDX example:

import MyComponent from './my-component'

# Title

<MyComponent />

Lorem ipsum dolor sit amet.

When processed with @mdx-js/loader this basically turns into the following JSX code:

import React from 'react'
import { MDXTag } from '@mdx-js/tag'
import MyComponent from './my-component'

export default ({ components }) => (
  <MDXTag name="wrapper" components={components}>
    <MDXTag name="h1" components={components}>
      Title
    </MDXTag>
    <MyComponent />
    <MDXTag name="p" components={components}>
      Lorem ipsum dolor sit amet.
    </MDXTag>
  </MDXTag>
)

This is what you actually import when you import a MDX file. Notice that the default export takes a components prop. You can either pass this directly or through context using MDXProvider, like in my previous example. The name prop of MDXTag maps to a component defined in the components prop. If components doesn't contain that key, it defaults to that tag name, i. e. "p" becomes <p> like in regular Markdown.

@brandonweiss
Copy link
Contributor Author

brandonweiss commented Jul 25, 2018

OK, I'm starting to get it, I think. So the components prop is a way of overriding an HTML tag to render a specific component instead. Most probably not to change the tag itself, but to apply some styling to it?

And the MDXProvider is necessary because otherwise there’d be no way to pass the components to the tags inside of an MDX file, because it’s Markdown/MDX (no render function and no props). Correct?

What would happen if you passed something other than a components prop to the MDXProvider? I guess nothing. There’d be no point in proxying other props because you wouldn’t be able to use them inside of MDX.

So the “Component customization“ section and the “MDXProvider” section of the README are… sort of related? I think one helps understand the other.

Would you be open to me improving those sections of the README?

@silvenon
Copy link
Contributor

silvenon commented Jul 25, 2018

Most probably not to change the tag itself, but to apply some styling to it?

Yeah. But another example could be to shift all headings. Imagine that you have a MDX file starting with h1, but in a certain context you need headings to start from h2. In that case you could push them all by one. But yes, generally you would use it for styling purposes.

And the MDXProvider is necessary

It's optional. It's only useful if you're frequently using the same components object. That's usually the case for blogs. You can pass components to each imported MDX file, like I showed in the first example. <One components={components}> means that every MDXTag in One will receive the same components object.

What would happen if you passed something other than a components prop to the MDXProvider? I guess nothing.

Yep, probably nothing.

Would you be open to me improving those sections of the README?

I'm not a maintainer, but go right ahead! Everybody loves documentation improvements. 👍

@silvenon
Copy link
Contributor

I'm closing this as a stale issue. This is the current documentation for MDXProvider: https://mdxjs.com/getting-started/#mdxprovider, and we're open to improvements.

wooorm added a commit that referenced this issue Jan 2, 2021
This PR moves most of the runtime to the compile time.

This issue has nothing to do with `@mdx-js/runtime`. It’s about
`@mdx-js/mdx` being compile time, and moving most work there, from the
“runtimes” `@mdx-js/react`, `@mdx-js/preact`, `@mdx-js/vue`.

Most of the runtime is undocumented features that allow amazing things,
but those are in my opinion *too magical*, more powerful than needed,
complex to reason about, and again: undocumented.
These features are added by overwriting an actual renderer (such as
react, preact, or vue). Doing so makes it hard to combine MDX with for
example Emotion or theme-ui, to opt into a new JSX transform when React
introduces one, to support other hyperscripts, or to add features such
as members (`<Foo.Bar />`). Removing these runtime features does what
MDX says in the readme: “**🔥 Blazingly blazing fast: MDX has no
runtime […]**”

This does remove the ability to overwrite *anything* at runtime. This
brings back the project to what is documented: users can still
overwrite markdown things (e.g., blockquotes) to become components and
pass components in at runtime without importing them. And it does still
allow undocumented parent-child combos (`blockquote.p`).

* Remove runtime renderers (`createElement`s hijacking) from
  `@mdx-js/react`, `@mdx-js/preact`, `@mdx-js/vue`
* Add `jsxRuntime` option to switch to the modern automatic JSX runtime
* Add `jsxImportSource` option to switch to a modern non-React JSX
  runtime
* Add `pragma` option to define a classic JSX pragma
* Add `pragmaFrag` option to define a classic JSX fragment
* Add `mdxProviderImportSource` option to load an optional runtime
  provider
* Add tests for automatic React JSX runtime
* Add tests for `@mdx-js/mdx` combined with `emotion`
* Add support and test members as “tag names” of elements
* Add support and test qualified names (namespaces) as “tag names” of
  elements
* Add tests for parent-child combos
* Add tests to assert explicit (inline) components precede over
  provided/given components
* Add tests for `mdxFragment: false` (runtime renderers w/o fragment
  support)
* Fix and test double quotes in attribute values

This PR removes the runtime renderers and related things such as the
`mdxType` and `parentName` props while keeping the `MDXProvider` in
tact.

This improves runtime performance, because all that runs at runtime is
plain vanilla React/preact/vue code.

This reduces the surface of the MDX API while being identical to what
is documented and hence to user expectations (except perhaps to some
power users).

This also makes it easier to support other renderers without having to
maintain projects like `@mdx-js/react`, `@mdx-js/preact`, `@mdx-js/vue`:
anything that can be used as a JSX pragma (including the [automatic
runtime](https://reactjs.org/blog/2020/09/22/introducing-the-new-jsx-transform.html))
is now supported.
A related benefit is that it’s easier to integrate with
[emotion](https://github.com/emotion-js/emotion/blob/master/packages/react/src/jsx.js#L7)
(including through `theme-ui`) and similar projects which also
overwrite the renderer: as it’s not possible to have two runtimes, they
were hard to combine; because with this PR MDX is no longer a renderer,
there’s no conflict anymore.

This is done by the compile time (`@mdx-js/mdx`) knowing about an
(**optional**) runtime for an `MDXProvider` (such as `@mdx-js/react`,
`@mdx-js/preact`). Importantly, it’s not required for other
hyperscript interfaces to have a provider: `MDXContent` exported from
a compiled MDX file *also* accepts components (it already did), and Vue
comes with component passing out of the box.

In short, the runtime looked like this:

```js
function mdx(thing, props, ...children) {
  const overwrites = getOverwritesSomeWay()
  return React.createElement(overwrites[props.mdxType] || thing, props, ...children)
}
```

And we had a compile time, which added that `mdxType` prop. So:

```mdx
<Youtube />
```

Became:

```js
const Youtube = () => throw new Error('Youtube is not loaded!')

<Youtube mdxType="Youtube" />
```

Which in plain JS looks like:

```js
const Youtube = () => throw new Error('Youtube is not loaded!')

React.createElement(Youtube, {mdxType: 'Youtube'})
```

Instead, this now compiles to:

```js
const {Youtube} = Object.assign({Youtube: () => throw new Error('Youtube is not loaded!')}, getOverwritesSomeWay())

React.createElement(Youtube)
```

The previous example shows what is sometimes called a “shortcode”: a
way to inject components as identifiers into the MDX file, which was
introduced in [MDX 1](https://mdxjs.com/blog/shortcodes)

A different use case for the runtime was overwriting “defaults”. This
is documented on the website as the “[Table of
components](https://mdxjs.com/table-of-components)”. This MDX:

```mdx
Hello, *world*!
```

Became:

```js
<p mdxType="p">Hello, <em mdxType="em">world</em>!</p>
```

This now compiles to:

```js
const overwrites = Object.assign({p: 'p', em: 'em'}, getOverwritesSomeWay())

<overwrites.p>Hello, <overwrites.em>world</overwrites.em>!</overwrites.p>
```

This MDX:

```mdx
export const Video = () => <Vimeo />

<Video />
```

Used like so:

```jsx
<MDXProvider components={{Video: () => <Youtube />}}>
  <Content />
</MDXProvider>
```

Would result in a `Youtube` component being rendered. It no longer
does. I see the previous behavior as a bug and hence this as a fix.

A subset of the above point is that:

```mdx
export default props => <main {...props} />

x
```

Used like so:

```jsx
<MDXProvider components={{wrapper: props => <article {...props} />}}>
  <Content />
</MDXProvider>
```

Would result in an `article` instead of the explicit `main`. It no
longer does. I see the previous behavior as a bug and hence this as a
fix.

(#821)

```mdx

<h2>World</h2>
```

Used like so:

```jsx
<MDXProvider components={{h2: () => <SomethingElse />}}>
  <Content />
</MDXProvider>
```

Would result in a `SomethingElse` for both. This PR **does not** change
that. But it could more easily be changed if we want to, because at
compile time we know whether something was a tag or not.

An undocumented feature of the current MDX runtime renderer is that
it’s possible to overwrite anything:

```mdx
<span />
```

Used like so:

```jsx
<MDXProvider components={{span: props => <b>{props.children}</b>}}>
  <Content />
</MDXProvider>
```

Would overwrite to become bold, even though it’s not documented
anywhere. This PR changes that: only allowed markdown “tag names” can
be changed (`p`, `li`, ...). **This list could be expanded.**

Another undocumented feature is that parent–child combos can be
overwritten. A `li` in an `ol` can be treated differently from one in
an `ul` by passing `'ol.li': () => <SomethingElse />`.

This PR no longer lets users “nest” arbitrary parent–child combos
except for `ol.li`, `ul.li`, and `blockquote.p`. **This list could
be expanded.**

It was not possible to use members (`<foo.bar />`, `<Foo.bar.baz />`,
<#953>) and supporting it previously
would be complex. This PR adds support for them.

Previously, `mdxType` and `parentName` attributes were added to all
elements. And a `components` prop was accepted on **all** elements to
change the provider. These are no longer passed and no longer accepted.
Lastly, `components`, `props` were in scope for all JSX tags defined in
the “markdown” section (not the import/exports) of each document.

This adds identifiers to the scope prefixed with double underscores:
`__provideComponents`, `__components`, and `__props`.

A single 1mb MDX file, about 20k lines and 135k words (basically 3
books). Heavy on the “markdown”, few tags, no import/exports.
322kb gzipped.

* v1: 2895.122856
* 2.0.0-next.8: 3187.4684129999996
* main: 4058.917152000001
* this pr: 4066.642403

* v1: raw: 1.5mb, gzip: 348kb
* 2.0.0-next.8: raw: 1.4mb, gzip: 347kb
* main: raw: 1.3mb, gzip: 342kb
* this pr: raw: 1.8mb, gzip: 353kb
* this pr, automatic runtime: raw: 1.7mb, gzip: 355kb

* v1: 321.761208
* 2.0.0-next.8: 321.79749599999997
* main: 162.412757
* this pr: 107.28038599999996
* this pr, automatic runtime: 123.73588899999999

This PR is much faster on giant markdown-esque documents during runtime.
The win over the current `main` branch is 34%, the win over the last
beta and v1 is 66%.

For output size, the raw value increases with this PR, which is because
the output is now `/*#__PURE__*/React.createElement(__components.span…)`
or `/*#__PURE__*/_jsx(__components.span…)`, instead of `mdx("span",
{mdxType: "span"…})`. The change is more repetition, as can be seen by
the roughly same gzip sizes.

That the build time of `main` and this PR is slower than v1 and the
last beta does surprise me a lot. I benchmarked earlier with 1000 small
simple MDX files, totalling 1mb, [where the results were the
inverse](#1399 (comment)). So
it looks like we have a problem with giant files. Still, this PR has no
effect on build time performance, because the results are the same as
currently on `main`.

This PR makes MDX faster, adds support for the modern automatic JSX
runtime, and makes it easier to combine with Emotion and similar
projects.

---

Some of what this PR does has been discussed over the years:

Related-to: GH-166.
Related-to: GH-197.
Related-to: GH-466 (very similar).
Related-to: GH-714.
Related-to: GH-938.
Related-to: GH-1327.

This PR solves some of the items outlined in these issues:

Related-to: GH-1152.
Related-to: #1014 (comment).

This PR solves:

Closes GH-591.
Closes GH-638.
Closes GH-785.
Closes GH-953.
Closes GH-1084.
Closes GH-1385.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Development

No branches or pull requests

2 participants