p tags get merged and className removed

[timsuchanek]

We're using mdx together with next.js' static export.
Everything works perfect, besides one problem.

Certain p tags get merged together when the javascript has been run on the static html.
The problem is that their className gets lost. As we use styled-components (would be the same problem with styled-jsx), the styling is broken at that point.
An example is this page: https://docs-beta.prisma.io/1.13/understand-prisma/prisma-basics-server-services-data-model-avrp/
At the very bottom, you'll see a paragraph with bad line-height.

If you disable javascript in the browser (in chrome dev tools settings -> preferences -> Debugger -> Disable JavaScript), you can see, that it's rendered correctly without javascript enabled, with an empty <p></p> tag before the p tag that will be broken.

I saw that mdx is using remark-squeeze-paragraphs which as far as I understand should fix exactly this problem. The actual remark-squeeze-paragraphs just removes not used p tags - so this should work.

Now the question is, where in the stack does this p tag merging happen and how can we fix it?

Thanks!

[emplums]

Having a similar issue, also using next.js. Except all my

tags are getting replaced with <path> elements 😭

[silvenon]

Hi! Is there any code reproducing this issue that I could take a look at? The link is now broken, btw.

[shawnbot]

@silvenon It's tricky because there are a lot of tools in the chain here: namely, Next.js, mdx, webpack, babel, and, terser. I think we've narrowed it down to the fact that one of the other libraries ends up with one of its functions mangled to have the name p (e.g. function p() { return React.createElement('path', ...) }. That function just happens to be a React component that renders a <path>. This is only happening for us in production, and we're trying to figure out how to disable the terser webpack plugin that might be producing (incorrectly, I think) the offending bundle:

https://github.com/zeit/next.js/blob/10a9178e3206270fd6e317df2e0a4dbffb460a6e/build/webpack.js#L87-L91

It may take a while for us to create a test case for this, but in the meantime can anyone think of any way that the remark transformation would produce React.createElement(p, ...) rather than React.createElement('p')? Is it possible that the component mapping for p elements is being clobbered by a function that's hoisted in the webpack bundle such that it would take precedence over it?

[silvenon]

Nothing like that comes to mind at the moment, but if you use JSX for tags natively supported by Markdown (<p>, <a> etc.), MDX has trouble with this (#246) and maybe in some combinations it might silently eat an element.

I'm looking forward to that test case once you manage to provide it.

[shawnbot]

No test case yet, but I did confirm that disabling mangling in next.config.js fixes the problem for us:

module.exports = {
  webpack(config, {dev}) {
    // only disable mangling in production
    if (dev) {
      return config
    }
    for (const plugin of config.plugins) {
      // duck type: is this an UglifyJS plugin?
      if (plugin.options && plugin.options.uglifyOptions) {
        console.warn('*** disabling mangling in UglifyJS plugin ***')
        plugin.options.uglifyOptions.compress = {keep_fnames: true}
        plugin.options.uglifyOptions.mangle.keep_fnames = true
      }
    }
    return config
  }
}

I've attempted to trace the source of whatever is calling our mangled function name in our production bundles, but I haven't figured it out yet. I'm 80% sure that the mangled icon function p() declaration is bleeding into MDX's scope somehow, and whatever is rendering paragraphs is calling that function instead.

[silvenon]

That's a pretty wild. It seems so unlikely that the string "p" turned into reference p at some point, even though it looks like a small thing, but I guess something like that is going on here. Meanwhile I'll make my own tests with mangling.

[johno]

I suspect that this is happening due to JS minification since we haven't been able to reproduce. If you are able to create a repro without code minifying please open an issue and we can keep digging!