fix(mdxish): parse <Table> as single token to prevent blank-line fragmentations by maximilianfalco · Pull Request #1371 · readmeio/markdown

maximilianfalco · 2026-03-12T09:17:40Z

	Fix RM-15613

🧰 Changes

CommonMark's HTML block type 6 (spec) matches <Table> case-insensitively against the table tag, which means it terminates at blank lines. Since <Table> JSX blocks always have blank lines between rows and sections, the mdxish pipeline was fragmenting them into multiple broken nodes.

Adds a custom micromark tokenizer (jsxTable) that grabs <Table> before the built-in htmlFlow parser and consumes everything through </Table> as a single token

An mdast-util-from-markdown extension turns that token into an html node
The existing mdxishTables transformer handles the rest by re-parsing it with remarkMdx into proper table nodes

Before	After

Demo

Screen.Recording.2026-03-16.at.19.30.56.mov

Warning

There still needs some work to be done on the monorepo side to fully support editing fragmented tables but for the moment, this PR ensures that we are rendering fragmented tables correctly. ie if you go back and forth in raw and rich mode in the new editor it breaks

This PR is essentially part 1 of solving RM-15601 and RM-15600

🧬 QA & Testing

Paste this in the demo app, it should render as ONE big table

<Table align={["left","left"]}>
  <thead>
    <tr>
      <th style={{ textAlign: "left" }}>
        Field
      </th>

      <th style={{ textAlign: "left" }}>
        Description
      </th>
    </tr>
  </thead>

  <tbody>
    <tr>
      <td style={{ textAlign: "left" }}>
        Fenced code block
      </td>

      <td style={{ textAlign: "left" }}>
        ```
        {
          "field": "ID",
          "type": "ASC"
        }
        ```
      </td>
    </tr>

    <tr>
      <td style={{ textAlign: "left" }}>
        Lists
      </td>

      <td style={{ textAlign: "left" }}>
        Oh no

        * no no no
        * no no no
        * Im so sorry
      </td>
    </tr>

    <tr>
      <td style={{ textAlign: "left" }}>
        Inline code with a pipe
      </td>

      <td style={{ textAlign: "left" }}>
        `foo | bar`
      </td>
    </tr>
  </tbody>
</Table>

and some edge case tests if you are interested

<Table>
  <tbody>
    <tr>
      <td>Literal closing tag as text</td>
      <td>\</Table> should not be separated</td>
    </tr>
  </tbody>
</Table>

---

<Table>
  <tbody>
    <tr>
      <td>Literal closing tag as text</td>
    </tr>
  </tbody>
</Table>

---

> <Table>
>   <tbody>
>     <tr><td>quoted</td></tr>
>   </tbody>
> </Table>
>
> text after quote

---

<Table>
  <thead>
  <tr><th>A</th></tr>
  </thead>
</Table>

---

<Table>
  <tbody>
    <tr>
      <td>Closing tag in code</td>
      <td>This `</Table>` should be auto-escaped.</td>
    </tr>
  </tbody>
</Table>

---

<Table>
  <tbody>
    <tr>
      <td>Closing tag in code</td>
    </tr>
  </tbody>

You should still see me!

---

<Table>
  <tbody>
    <tr>
      <td>
        <Table>
          <tbody>
            <tr>
              <td>Nested Table</td>
            </tr>
          </tbody>
        </Table>
      </td>
      <td>Hi</td>
    </tr>
  </tbody>
</Table>

pass

…rm-15613-properly-parse-and-render-fragmented-table-elements

__tests__/parsers/tables.test.ts

lib/micromark/jsx-table/syntax.ts

lib/stripComments.ts

processor/transform/mdxish/mdxish-tables.ts

package.json

__tests__/parsers/tables.test.ts

maximilianfalco · 2026-03-13T06:07:45Z

okay im trying something new in this commit 753a8e4

i realised now that the tokenizer can ensure the entire<Table> comes in one singular node, we can make use of remarkMdx and all its benefits to parse cell contents instead relying on the current lightweight (remarkParse + remarkGfm only) parser. hopefully this'll allow use to have parity with both RDMD and RMDX.

processor/transform/mdxish/mdxish-tables.ts

maximilianfalco · 2026-03-16T06:19:25Z

with the new tokenizer, the entire <Table> ... </Table> should always come in a single block which means that the transformer no longer has to handle multiple different possible shapes hence why we are able to remove a lot of stuff in mdxish-tables

the isMDXElement visit pass and mdxishComponentBlocks import are dead code since the tokenizer captures the table before mdxishComponentBlocks ever sees it
handleTableInNode + separate html/raw visits are no longer needed since the tokenizer guarantees a single html node, so they collapse into one visit(tree, 'html', ...)
replaced mdCellProcessor / parseMarkdown with tableNodeProcessor which uses remarkMdx instead of lightweight remarkParse + remarkGfm, so cells now support full mdxish syntax (callouts, gemoji, components, etc)

EDIT (some follow ups):

tokenizer now handles backslash-escaped \</Table> in cell content without prematurely closing the block
tables without <thead> are kept as JSX instead of converting to mdast (mdast always treats the first row as a header, so body rows would get promoted)
single-line <tr><th>...</th></tr> now renders correctly
paragraph wrappers and whitespace-only text nodes produced by remarkMdx are cleaned up in the JSX path to prevent empty <p>/<br> gaps

rafegoldberg

This is working well, save for a couple issues I noticed:

The second one is really more of an edge case, but the first one I think we actually need to fix before this gets merged?

run the start script to view these links ↩ ↩²

maximilianfalco · 2026-03-17T03:22:46Z

lib/constants.ts

+ *
+ * @see https://github.com/syntax-tree/mdast#flowcontent
+ */
+export const FLOW_TYPES = new Set([


exposing a list of FLOW_TYPES here for consumers to use, I was deliberating to make a new Set over in the monorepo but opted to just put this as an import so we can a single SOT

This is to help with differentiating when to use GFM vs JSX table syntax in the editor serialization right?

yea, tldr, if the table contains any FLOW types (except html) then we serialize to JSX syntax to keep it rich (ie to allow code blocks)

eaglethrost · 2026-03-17T04:57:17Z

__tests__/parsers/tables.test.ts

+      expect(cells).toHaveLength(2);
+    });
+
+    it('does not swallow content after an unclosed Table', () => {


Do we also need to check there's no table nodes in the tree?

you should still be able to see tables, this is on par with RDMD

<Table> <tbody> <tr> <td>Closing tag in code</td> </tr> </tbody> You should still see me!

## Version 13.6.1 ### 🛠 Fixes & Updates * **mdxish:** parse <Table> as single token to prevent blank-line fragmentations ([#1371](#1371)) ([ce73b7f](ce73b7f)) * **mdxish:** variables in callout titles causing it to crash ([#1378](#1378)) ([4006247](4006247)), closes [#1362](#1362)

rafegoldberg · 2026-03-17T18:31:51Z

This PR was released!

🚀 Changes included in v13.6.1

maximilianfalco added 6 commits March 12, 2026 19:52

feat: introduce micromark tokenizer for JSX tables

ca13605

refactor: simplify the mdxish-table transformer to a single html visit

e9eb136

pass

feat: integrate new tokenizer into the mdxish pipeline

af17b0e

chore: add some tests

c7fb6ea

Merge branch 'next' of https://github.com/readmeio/markdown into falco/…

fb98eaf

…rm-15613-properly-parse-and-render-fragmented-table-elements

bundlewatch

468b042

maximilianfalco requested review from Jadenzzz and eaglethrost March 12, 2026 22:09

add this tokenizer to stripComments

ad7a7ee