fix(mdxish): preserve multi-paragraph table cells by eaglethrost · Pull Request #1445 · readmeio/markdown

eaglethrost · 2026-04-23T13:01:29Z

🎫 Resolve RM-16217

🎯 What does this PR do?

Magic-block `[block:parameters]` tables (and actually JSX Tables as well) with cells containing paragraph breaks (`\n\n`) lose those breaks after a save through the MdxishEditor. A cell rendered as three paragraphs collapses into a single concatenated line (`para1para2para3`).

There are 2 issues causing this:

The magic block table was getting serialised to GFM tables, which don't actually support multiline paragraphs in a table cell
Even then, sometimes the newline / paragraph info got lost in the mdxish pipeline. This is because of a logic in the table transformer mdxish-tables.ts to flatten and unwrap paragraph nodes in a table cell, hence it'll become just an array of text nodes, which bunches them to one line

Fixes:

tables-to-jsx: treat `>1` paragraph children in a cell as flow content, forcing JSX `` output instead of GFM
jsx-to-mdast / mdxish-tables: only unwrap a paragraph child when it is the cell's sole paragraph. Multiple paragraphs stay intact. I think we still need to unwrap if there's only 1 paragraph as it might cause some visual issues

🧪 QA tips

Paste the following into a doc with the MdxishEditor, save, and confirm the three paragraphs are preserved (not collapsed to `Line 1Line 2Line 3`):

[block:parameters]
{
  "data": {
    "h-0": "Item",
    "0-0": "Line 1 \n\nLine 2 \n\nLine 3"
  },
  "cols": 1,
  "rows": 1,
  "align": ["left"]
}
[/block]

Also verify a JSX <Table> with blank-line-separated content in a cell round-trips cleanly:

<Table align={["left"]}>
  <thead>
    <tr>
      <th>Item</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>
        Line 1

        Line 2

        Line 3
      </td>
    </tr>
  </tbody>
</Table>

Magic-block table cell with `\n\n` preserves paragraphs after save
JSX <Table> cell with blank lines preserves paragraphs after save
Single-paragraph cells still render as plain GFM tables (no regression)

📸 Screenshot or Loom

BEFORE (Notice how the new lines get lost after the round-trip):

Screen.Recording.2026-04-23.at.11.25.04.pm.mov

AFTER (Serialising to JSX table and preserving newlines):

Screen.Recording.2026-04-23.at.11.25.34.pm.mov

NOTE that in the editor, the paragraph gaps are not exactly the same as in view mode. If we want to update that to be the exact same, we'd need to look into the editor side. In spite of that, the round-trip works and the view mode still renders with the correct paragraph gaps.

…te lines

eaglethrost · 2026-04-23T13:11:29Z

+        const paragraphCount = cell.children.filter(c => c.type === 'paragraph').length;
+        const parsedChildren =
+          paragraphCount === 1
+            ? cell.children.flatMap((parsedNode: Node) => {


This logic is kind of a duplicate to the one in mdxish-tables. I'm thinking of moving the table-related transformers under 1 subfolder under transformers/ so that it can host its own util file and share code.

Yeah I agree with doing this. Also fine with putting an unwrapSoleParagraph (or whatever name makes sense) function into processor/utils. But we definitely don't want two duplicates that could drift if one gets updated without the other.

eaglethrost · 2026-04-23T14:01:15Z

As I've alluded to in the note below the loom, even with this fix, the paragraphs in the table cell in the editor are only 1 line apart & different from the view mode. I don't think there's any more we can do in the engine though, and it would be a fix in the editor.

Unfortunately I think this part isn't a straightforward fix, one thing we could do is to add break tags to the tree to make sure the editor matches the view mode, but it will add it in the raw mode as well which I'm not sure if we want to do more of than we already do. Will think of that more.

Editor	View

eaglethrost · 2026-04-23T14:04:44Z

The as Table casts are to fix type errors

This is because MDAST Table doesn't expect paragraphs in table cell children right?

Can we instead type the expected shape and use it here? I also guess we'd need to update a table transformer that is also casting unsafely.

We've got this types file that already holds transitional node shapes we could put a new Table & TableCell type in https://github.com/readmeio/markdown/blob/next/processor/transform/mdxish/types.ts

Yeah because the MDAST table cell children is typed as PhrasingContent[] which doesn't accept flow level nodes such as Paragraphs, Lists, etc.

Yes we can definitely extend the table types! I've had a thought about it, I think we should do it in a follow up so we can integrate the types more holistically and have some room for discussions as well. Will create one soon.

eaglethrost · 2026-04-23T14:05:01Z

@@ -251,6 +257,7 @@ const mdxishTables = (): Transform => tree => {
          // we let it get handled naturally
          return EXIT;
        }
+        return undefined;


Lint error fix, shouldn't affect anything

kevinports

Approving with a few comments that'd be good to address before merging.

As I've alluded to in the note below the loom, even with this fix, the paragraphs in the table cell in the editor are only 1 line apart & different from the view mode. I don't think there's any more we can do in the engine though, and it would be a fix in the editor.

Unfortunately I think this part isn't a straightforward fix, one thing we could do is to add break tags to the tree to make sure the editor matches the view mode, but it will add it in the raw mode as well which I'm not sure if we want to do more of than we already do. Will think of that more.

Yeah it'd definitely be nice to clean this up. Could it just be a CSS styling difference between view vs edit rendering?

kevinports · 2026-04-23T15:16:28Z

+        const paragraphCount = cell.children.filter(c => c.type === 'paragraph').length;
+        const parsedChildren =
+          paragraphCount === 1
+            ? cell.children.flatMap((parsedNode: Node) => {


Yeah I agree with doing this. Also fine with putting an unwrapSoleParagraph (or whatever name makes sense) function into processor/utils. But we definitely don't want two duplicates that could drift if one gets updated without the other.

kevinports · 2026-04-23T15:30:02Z

This is because MDAST Table doesn't expect paragraphs in table cell children right?

Can we instead type the expected shape and use it here? I also guess we'd need to update a table transformer that is also casting unsafely.

We've got this types file that already holds transitional node shapes we could put a new Table & TableCell type in https://github.com/readmeio/markdown/blob/next/processor/transform/mdxish/types.ts

## Version 14.1.1 ### 🛠 Fixes & Updates * DRY up some test helpers ([#1437](#1437)) ([f761248](f761248)) * **mdxish:** preserve multi-paragraph table cells ([#1445](#1445)) ([c821b2d](c821b2d)) * use magic block tokenizer when parsing JSX component children ([#1447](#1447)) ([51b22a9](51b22a9)), closes [#1429](#1429)

rafegoldberg · 2026-04-24T21:35:05Z

This PR was released!

🚀 Changes included in v14.1.1

eaglethrost added 2 commits April 23, 2026 22:54

fix: dont unwrap multiline paragraphs in table cells to retain separa…

001a817

…te lines

fix: paragraphs in cells should serialize to jsx table

3f17522

eaglethrost changed the title ~~fix(mdxish): preserve multi-paragraph table cells across round-trip (RM-16217)~~ fix(mdxish): preserve multi-paragraph table cells Apr 23, 2026

eaglethrost commented Apr 23, 2026

View reviewed changes

eaglethrost requested review from dannobytes, kevinports and maximilianfalco April 23, 2026 13:32

eaglethrost commented Apr 23, 2026

View reviewed changes

kevinports approved these changes Apr 23, 2026

View reviewed changes

eaglethrost added 3 commits April 24, 2026 13:36

refactor: move table transformers

5b417b9

style: move unwrap

4bbcb1a

test: single paragraph

e9bbdfb

eaglethrost merged commit c821b2d into next Apr 24, 2026
7 checks passed

eaglethrost deleted the dimas/rm-16217-editor-drops-n-new-lines-from-magic-block-table-cells-on branch April 24, 2026 06:42

eaglethrost mentioned this pull request Apr 24, 2026

refactor(mdxish): extend table types to reflect flow-content cells #1446

Open

2 tasks

rafegoldberg added the released label Apr 24, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(mdxish): preserve multi-paragraph table cells#1445

fix(mdxish): preserve multi-paragraph table cells#1445
eaglethrost merged 5 commits into
nextfrom
dimas/rm-16217-editor-drops-n-new-lines-from-magic-block-table-cells-on

eaglethrost commented Apr 23, 2026 •

edited

Loading

Uh oh!

eaglethrost Apr 23, 2026

Uh oh!

kevinports Apr 23, 2026

Uh oh!

eaglethrost commented Apr 23, 2026 •

edited

Loading

Uh oh!

eaglethrost Apr 23, 2026

Uh oh!

kevinports Apr 23, 2026

Uh oh!

eaglethrost Apr 24, 2026

Uh oh!

eaglethrost Apr 23, 2026 •

edited

Loading

Uh oh!

kevinports left a comment

Uh oh!

kevinports Apr 23, 2026

Uh oh!

kevinports Apr 23, 2026

Uh oh!

Uh oh!

rafegoldberg commented Apr 24, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

eaglethrost commented Apr 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🎯 What does this PR do?

🧪 QA tips

📸 Screenshot or Loom

Uh oh!

eaglethrost Apr 23, 2026

Choose a reason for hiding this comment

Uh oh!

kevinports Apr 23, 2026

Choose a reason for hiding this comment

Uh oh!

eaglethrost commented Apr 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

eaglethrost Apr 23, 2026

Choose a reason for hiding this comment

Uh oh!

kevinports Apr 23, 2026

Choose a reason for hiding this comment

Uh oh!

eaglethrost Apr 24, 2026

Choose a reason for hiding this comment

Uh oh!

eaglethrost Apr 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kevinports left a comment

Choose a reason for hiding this comment

Uh oh!

kevinports Apr 23, 2026

Choose a reason for hiding this comment

Uh oh!

kevinports Apr 23, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

rafegoldberg commented Apr 24, 2026

This PR was released!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

eaglethrost commented Apr 23, 2026 •

edited

Loading

eaglethrost commented Apr 23, 2026 •

edited

Loading

eaglethrost Apr 23, 2026 •

edited

Loading