[lexical-mdast][lexical-playground][lexical-website] Bug Fix: Write line breaks in Markdown table cells as <br>

[etrepum]

Description

A GFM table cell holds a single line, so line breaks in a cell need the <br> convention that GitHub and most renderers accept there.

• The playground's TABLE Markdown transformer wrote a literal \n for each newline in a cell (one\n\ntwo), which no renderer reads. It now writes <br>, escapes | in cell text, reads <br>, <br/> and <br /> back (the legacy \n too) except inside code spans and after a backslash escape, escapes a <br> typed as text, writes each line of a fenced code block with its own <br> (so blank lines survive) and backslash-escapes any \, | or <br> in the code so it reads back verbatim, keeps bold, italic, strikethrough and links whole when a <br> falls inside them, gives each pipe in inline code a backslash of its own and writes code with a backslash right before a pipe, which a GFM cell can't hold as a code span, as <code> with character references, splits rows only on unescaped pipes (a pipe after an odd run of backslashes is escaped, as in micromark: \| and \\\| are content, \\| splits), trims cell padding on import, writes the space or tab next to a line break as a character reference so it isn't trimmed as padding, writes an empty paragraph as an empty line (a<br><br>b), reads a row of empty cells (| | |) as a row rather than a delimiter row, and reads a cell with many line breaks in linear time. A hard line break's \ marker no longer lands in front of the <br> (a\<br>b read back as an escaped <), a single-column table with a header row now gets its delimiter row, and | --- | with no table above it stays text instead of becoming an empty paragraph.
• @lexical/markdown read the backslash escapes in formatted text twice, so **a\\\*b** came back as a*b; each escape is now read once.
• @lexical/markdown dropped backslashes from a code span that was all of a text node's text (`\*` read back as *, ` ` as a no-break space), though code spans take no escapes; it now leaves code-formatted text alone. This is the fix in [lexical-markdown] Bug Fix: Preserve literal inline-code content during import #9333; the one case its tests check that this PR's didn't (text after a leading code span still reads its escapes) is included here.
• @lexical/markdown kept a line with its matched prefix cut off when an element transformer cancelled its import by returning false. It now restores the whole line, unless the transformer changed or moved the text before cancelling, in which case its changes stay as before.
• @lexical/mdast's MdastTableExtension exported a cell's paragraphs and line breaks as spaces, so the lines were lost, and fused the items of a list in a cell (xy). It now writes each paragraph and line break as its own <br>-separated line, and imports <br> as a paragraph boundary (nested in formatting, as a line break). Without this, <br> imported as literal text and exported as \<br>.
• When the editor also has MdastHtmlExtension (a peer dependency of MdastTableExtension), lists, task lists, headings, quotes, code blocks and rules in a cell are written as one line of HTML around their Markdown content, such as intro<ol start="3"><li>one<ul><li>nested</li></ul></li></ol>, which GFM allows in a cell, and read back as the same blocks. Without it, they are flattened into <br>-separated lines. Raw HTML in a cell is written on one line: a newline in its text or a quoted attribute value becomes 
, so whitespace that <pre> or CSS keeps survives, a newline between attributes or in a comment becomes a space, the text of <textarea>, <script> and the like is read to its end tag, and every pipe becomes |, so <span title="a|b"> doesn't split the row. An htmlBlock (rawHtmlBlock) in a cell is written as one line of its HTML around its Markdown children.
• Column alignment in @lexical/mdast was stored as one array on the table, so deleting or inserting a column shifted every alignment after it, and the editor never showed it. It is now the element format of each cell in the column, which renders as text-align and which FORMAT_ELEMENT_COMMAND already sets on a cell selection; export also reads the format a cell's paragraphs share, and since GFM can't align one cell, a column takes the alignment a strict majority of its cells have, or else that of its cell in the table's header row, so inserted rows don't clear it. An existing document's old table-level state is moved onto its cells that have no format of their own when the table loads and then cleared, so it can't override a column whose alignment is later cleared or moved. A cell that already has a format keeps it, including justify, start and end, which GFM can't write; the docs say so and tests cover it.
• The mdast-editor dev example gets table editing: TableExtension (with the features GFM can't express turned off), Insert > Table, a Table menu for adding and removing rows and columns and aligning columns, table styles, and a Tables section in the demo document.
• The mdast-editor dev example also clears its footnotes section when everything else in the document is deleted, and styles the block cursor (theme.blockCursor) Lexical draws between adjacent tables or decorators.

Closes #9323

Test plan

Before

$ vitest run --project unit packages/lexical-mdast/src/__tests__/unit/MdastImportExport.test.ts packages/lexical-playground/__tests__/unit/MarkdownTransformers.test.ts
     × writes line breaks in a cell as <br> and escapes pipes
     × reads <br>, escaped pipes and the legacy \n back
       × keeps each column its alignment when a column is removed
       × gives an inserted column no alignment
       × joins multi-paragraph cells instead of fusing their text
       × writes line breaks in a cell as <br>, never a newline
       × reads <br> in a cell as a paragraph boundary
AssertionError: expected '| a | b |\n| --- | --- |\n| one\n\ntw…' to be '| a | b |\n| --- | --- |\n| one<br>tw…'
AssertionError: expected '| b  |\n| :- |\n| 2  |' to be '|  b |\n| -: |\n|  2 |'
AssertionError: expected 'a b' to be 'a<br>b'
AssertionError: expected [ 'x <br/> y<BR /><br>z<br>w' ] to deeply equal [ 'x', 'y', '', 'z\nw' ]
      Tests  7 failed | 103 passed (110)

After

$ vitest run --project unit packages/lexical-mdast packages/lexical-markdown packages/lexical-playground packages/lexical-code-core
      Tests  1471 passed (1471)
$ vitest run --project browser dev-examples/mdast-editor packages/lexical-mdast
      Tests  44 passed (44)
$ vitest run --project browser dev-examples/mdast-editor packages/lexical-mdast packages/lexical-playground
      Tests  187 passed | 1 skipped (188)
$ tsc -p .   # exit 0

The new browser test dev-examples/mdast-editor/src/__tests__/browser/TableEdit.test.ts covers the Table menu's commands, pressing Enter in a cell, which exports | 1 | 2<br>3 |, and a list and code block in a cell round-tripping as HTML. MdastTableHtml.test.ts covers the HTML cell format, and the playground and markdown tests cover the issue comments' cases. Browser tests ran in Chromium only; Firefox and WebKit were not exercised.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
lexical	Ready	Preview	Oct 8, 2026 1:12am UTC
lexical-playground	Ready	Preview	Oct 8, 2026 1:12am UTC