feat(math): emit data-mx-maths on send for cross-client LaTeX

Math rendering already shipped (both the $…$ shorthand and the spec
data-mx-maths form render incoming). But the composer emitted no math, so a
Lotus user's $E=mc^2$ went out as raw text — it rendered on Lotus (both sides
parse $…$) but showed as literal text on Element and other clients.

toMatrixCustomHTML now converts $…$/$$…$$ to
  <span|div data-mx-maths="LATEX"><code>LATEX</code></span|div>
(spec CS-API §11.5), reusing the existing splitMathSegments parser. Math is
extracted BEFORE markdown so LaTeX (_, *, \, {}) isn't mangled, and the emitted
span survives the block-markdown pass via the existing ignoreHTMLParseInlineMD
HTML-tag guard. A new allowMath opt threads through the top-level call sites;
code-line/code-block paths use empty opts so math is off inside code. The plain
body keeps literal $…$ as the fallback.

Scope: inline $…$ + single-line $$…$$. Multi-line block $$ (spans editor
paragraph nodes) deferred — still renders on Lotus via the plain-body path.

New output.test.ts (8 cases): span/div emission, escaping, markdown-bypass,
currency non-match, code-mark + code-block exclusion. 738 tests pass.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
2026-07-07 12:46:25 -04:00
parent 7ad948e26c
commit 33cb103abb
4 changed files with 99 additions and 0 deletions
+68
View File
@@ -0,0 +1,68 @@
import { test } from 'node:test';
import assert from 'node:assert/strict';
import { Descendant } from 'slate';
import { toMatrixCustomHTML } from './output';
import { BlockType } from './types';
// Loose Slate node builders for the test.
const txt = (text: string, marks: Record<string, unknown> = {}) =>
({ text, ...marks }) as unknown as Descendant;
const el = (type: BlockType, children: unknown[]) =>
({ type, children }) as unknown as Descendant;
const OPTS = {
allowTextFormatting: true,
allowInlineMarkdown: false,
allowBlockMarkdown: false,
allowMath: true,
};
test('inline $…$ → data-mx-maths span, surrounding prose preserved', () => {
assert.equal(
toMatrixCustomHTML(txt('a $x^2$ b'), OPTS),
'a <span data-mx-maths="x^2"><code>x^2</code></span> b',
);
});
test('single-line block $$…$$ → data-mx-maths div', () => {
assert.equal(
toMatrixCustomHTML(txt('$$E=mc^2$$'), OPTS),
'<div data-mx-maths="E=mc^2"><code>E=mc^2</code></div>',
);
});
test('LaTeX special chars are escaped in attribute + fallback', () => {
assert.equal(
toMatrixCustomHTML(txt('$a<b$'), OPTS),
'<span data-mx-maths="a&lt;b"><code>a&lt;b</code></span>',
);
});
test('math bypasses markdown — underscores are not italicised', () => {
// Without pre-markdown extraction, `_` would become <em>.
assert.equal(
toMatrixCustomHTML(txt('$a_b$'), { ...OPTS, allowInlineMarkdown: true }),
'<span data-mx-maths="a_b"><code>a_b</code></span>',
);
});
test('currency ($5 and $10) stays literal', () => {
assert.equal(toMatrixCustomHTML(txt('$5 and $10'), OPTS), '$5 and $10');
});
test('no math conversion inside an inline code mark', () => {
const out = toMatrixCustomHTML(txt('$x$', { code: true }), OPTS);
assert.ok(!out.includes('data-mx-maths'));
assert.ok(out.includes('<code>$x$</code>'));
});
test('no math conversion inside a code block', () => {
const block = el(BlockType.CodeBlock, [el(BlockType.CodeLine, [txt('$x$')])]);
const out = toMatrixCustomHTML(block, OPTS);
assert.ok(!out.includes('data-mx-maths'));
assert.ok(out.includes('$x$'));
});
test('a message with no math is unchanged', () => {
assert.equal(toMatrixCustomHTML(txt('just hello'), OPTS), 'just hello');
});
+28
View File
@@ -12,14 +12,42 @@ import {
import { findAndReplace } from '../../utils/findAndReplace';
import { sanitizeForRegex } from '../../utils/regex';
import { isUserId } from '../../utils/matrix';
import { splitMathSegments } from '../../utils/mathParse';
export type OutputOptions = {
allowTextFormatting?: boolean;
allowInlineMarkdown?: boolean;
allowBlockMarkdown?: boolean;
allowMath?: boolean;
};
// Spec `data-mx-maths` markup (CS-API §11.5): the attribute holds the LaTeX; the
// <code> child is the fallback for clients that don't render math. sanitizeText
// escapes & < > " ' — correct for both the attribute value and the <code> text.
const mathToCustomHtml = (latex: string, block: boolean): string => {
const esc = sanitizeText(latex);
const tag = block ? 'div' : 'span';
return `<${tag} data-mx-maths="${esc}"><code>${esc}</code></${tag}>`;
};
const textToCustomHtml = (node: Text, opts: OutputOptions): string => {
// Convert `$…$`/`$$…$$` to `data-mx-maths` so math renders on other clients.
// Extracted BEFORE markdown so LaTeX (`_`, `*`, `\`, `{}`) isn't mangled; never
// applied inside inline code. Non-math text recurses with allowMath off so it
// still gets the normal marks + inline-markdown treatment.
if (opts.allowMath && !node.code) {
const segments = splitMathSegments(node.text);
if (segments.some((seg) => seg.type !== 'text')) {
return segments
.map((seg) =>
seg.type === 'text'
? textToCustomHtml({ ...node, text: seg.value }, { ...opts, allowMath: false })
: mathToCustomHtml(seg.value, seg.type === 'block')
)
.join('');
}
}
let string = sanitizeText(node.text);
if (opts.allowTextFormatting) {
if (node.bold) string = `<strong>${string}</strong>`;
+2
View File
@@ -514,6 +514,7 @@ export const RoomInput = forwardRef<HTMLDivElement, RoomInputProps>(
allowTextFormatting: true,
allowBlockMarkdown: isMarkdown,
allowInlineMarkdown: isMarkdown,
allowMath: true,
}),
);
let msgType = MsgType.Text;
@@ -616,6 +617,7 @@ export const RoomInput = forwardRef<HTMLDivElement, RoomInputProps>(
allowTextFormatting: true,
allowBlockMarkdown: isMarkdown,
allowInlineMarkdown: isMarkdown,
allowMath: true,
}),
);
if (plainText === '') return null;
@@ -117,6 +117,7 @@ export const MessageEditor = as<'div', MessageEditorProps>(
allowTextFormatting: true,
allowBlockMarkdown: isMarkdown,
allowInlineMarkdown: isMarkdown,
allowMath: true,
}),
);