Fix small text not being escapable and triggering in code blocks.

.changeset/fix-subtext-codeblocks.md

@@ -0,0 +1,5 @@
+---
+default: patch
+---
+
+Fix small text being parsed in code blocks and not being escapeable.

src/app/plugins/markdown/bidirectional.test.ts

@@ -107,6 +107,16 @@ describe('bidirectional round-trip', () => {
     expect(result).toContain('||hidden message||');
   });
 
+  it('round-trips literal line-start -# (escaped) in a paragraph', () => {
+    const markdown = '\\-# not small text';
+    const html = markdownToHtml(markdown);
+    expect(html).not.toContain('<sub');
+    const injected = injectDataMd(html);
+    const result = htmlToMarkdown(injected);
+    expect(result).toContain('\\-#');
+    expect(result).toContain('not small text');
+  });
+
   it('round-trips inline math', () => {
     const markdown = '$E = mc^2$';
     const html = markdownToHtml(markdown);

src/app/plugins/markdown/extensions/matrix-math.ts

@@ -3,6 +3,22 @@ import type { TokenizerExtension, RendererExtension } from 'marked';
 /** Private-use char so math extensions do not match `$` / `$$` inside code spans. Not U+E000–U+E002 (emoticon placeholders). {@link shieldDollarRunsForMarked} uses U+E021–U+E022. */
 export const MATH_CODE_DOLLAR_MASK = '\uE020';
 
+/**
+ * Replaces the `-` of line-start `-# …` inside markdown code so the Matrix subscript block
+ * extension does not match before marked's `fences` rule (custom block extensions run first).
+ * {@link unmaskSubscriptCodeLinePlaceholders} restores output HTML.
+ */
+export const SUBSCRIPT_CODE_LINE_MASK = '\uE023';
+
+function maskSubscriptLineStartsInCodeInner(inner: string): string {
+  return inner.replace(/(^|\n)-#( +)/g, `$1${SUBSCRIPT_CODE_LINE_MASK}#$2`);
+}
+
+/** Applies {@link MATH_CODE_DOLLAR_MASK} and subscript masking inside a fence or inline-code region. */
+function maskMathAndSubscriptInCodeInner(inner: string): string {
+  return maskSubscriptLineStartsInCodeInner(inner.replace(/\$/g, MATH_CODE_DOLLAR_MASK));
+}
+
 function findSameLineFenceClose(md: string, from: number, tick: string, minLen: number): number {
   let j = from;
   while (j < md.length && md[j] !== '\n') {
@@ -62,11 +78,11 @@ function tryConsumeFence(md: string, i: number): { text: string; end: number } |
     const close = findMultilineFenceEnd(md, contentStart, tick, openLen);
     if (!close) {
       const inner = md.slice(contentStart, md.length);
-      const masked = inner.replace(/\$/g, MATH_CODE_DOLLAR_MASK);
+      const masked = maskMathAndSubscriptInCodeInner(inner);
       return { text: md.slice(i, contentStart) + masked, end: md.length };
     }
     const inner = md.slice(contentStart, close.contentEnd);
-    const maskedInner = inner.replace(/\$/g, MATH_CODE_DOLLAR_MASK);
+    const maskedInner = maskMathAndSubscriptInCodeInner(inner);
     return {
       text: md.slice(i, contentStart) + maskedInner + md.slice(close.contentEnd, close.blockEnd),
       end: close.blockEnd,
@@ -80,7 +96,7 @@ function tryConsumeFence(md: string, i: number): { text: string; end: number } |
   while (closeIdx + closeRun < md.length && md[closeIdx + closeRun] === tick) closeRun++;
 
   const inner = md.slice(afterOpen, closeIdx);
-  const maskedInner = inner.replace(/\$/g, MATH_CODE_DOLLAR_MASK);
+  const maskedInner = maskMathAndSubscriptInCodeInner(inner);
   return {
     text: md.slice(i, afterOpen) + maskedInner + md.slice(closeIdx, closeIdx + closeRun),
     end: closeIdx + closeRun,
@@ -99,7 +115,7 @@ function tryConsumeInlineCode(md: string, i: number): { text: string; end: numbe
       while (j + cr < md.length && md[j + cr] === '`') cr++;
       if (cr === run) {
         const inner = md.slice(contentStart, j);
-        const maskedInner = inner.replace(/\$/g, MATH_CODE_DOLLAR_MASK);
+        const maskedInner = maskMathAndSubscriptInCodeInner(inner);
         return {
           text: md.slice(i, contentStart) + maskedInner + md.slice(j, j + run),
           end: j + run,
@@ -155,6 +171,10 @@ export function unmaskMathCodeDollarPlaceholders(html: string): string {
   return html.replaceAll(MATH_CODE_DOLLAR_MASK, '$');
 }
 
+export function unmaskSubscriptCodeLinePlaceholders(html: string): string {
+  return html.replaceAll(`${SUBSCRIPT_CODE_LINE_MASK}#`, '-#');
+}
+
 const MARKED_MATH_BLOCK_SHIELD = '\uE021';
 const MARKED_MATH_BLOCK_SHIELD_END = '\uE022';
 

src/app/plugins/markdown/extensions/matrix-subscript.ts

@@ -4,25 +4,24 @@ import type { TokenizerExtension, RendererExtension, Tokens } from 'marked';
 export const matrixSubscriptExtension = {
   name: 'subscript',
   level: 'block',
-  start(src: string) {
-    return src.indexOf('-#');
-  },
   tokenizer(
-    this: { lexer: { inlineTokens: (t: string, tokens: Tokens.Generic[]) => void } },
+    this: {
+      lexer: { inlineTokens: (t: string, tokens: Tokens.Generic[]) => void };
+    },
     src: string
   ) {
-    const match = /^-# +(.+)/.exec(src);
-    if (match) {
-      const token = {
-        type: 'subscript',
-        raw: match[0],
-        text: match[1],
-        tokens: [] as Tokens.Generic[],
-      };
-      this.lexer.inlineTokens(token.text!, token.tokens);
-      return token;
+    const match = /^-# +([^\n]+)/.exec(src);
+    if (!match) {
+      return undefined;
     }
-    return undefined;
+    const token = {
+      type: 'subscript',
+      raw: match[0],
+      text: match[1],
+      tokens: [] as Tokens.Generic[],
+    };
+    this.lexer.inlineTokens(token.text!, token.tokens);
+    return token;
   },
   renderer(
     this: { parser: { parseInline: (tokens: Tokens.Generic[]) => string } },

src/app/plugins/markdown/markdownToHtml.test.ts

@@ -1,4 +1,5 @@
 import { describe, expect, it } from 'vitest';
+import { htmlToMarkdown } from './htmlToMarkdown';
 import { markdownToHtml } from './markdownToHtml';
 
 describe('markdownToHtml', () => {
@@ -89,6 +90,41 @@ describe('markdownToHtml', () => {
     expect(result).toContain('$$test$$');
   });
 
+  it('converts -# small/sub syntax outside code', () => {
+    const result = markdownToHtml('-# caption');
+    expect(result).toContain('<sub');
+    expect(result).toContain('data-md="-#"');
+    expect(result).toContain('caption');
+  });
+
+  it('does not parse -# inside fenced code as subscript', () => {
+    expect(markdownToHtml('```\n-# not sub\n```')).not.toContain('<sub');
+    expect(markdownToHtml('```\n-# not sub\n```')).toContain('-# not sub');
+  });
+
+  it('does not parse -# inside inline code as subscript', () => {
+    expect(markdownToHtml('`-# lit`')).not.toContain('<sub');
+    expect(markdownToHtml('`-# lit`')).toContain('-# lit');
+  });
+
+  it('parses -# as single-line only so fenced code below stays code', () => {
+    const html = markdownToHtml('-# caption\n```\nfenced\n```');
+    expect(html).toContain('caption');
+    expect(html).toContain('<pre>');
+    expect(html).toContain('fenced');
+  });
+
+  it('does not parse escaped \\-# as small/sub', () => {
+    const result = markdownToHtml('\\-# literal caption');
+    expect(result).not.toContain('<sub');
+    expect(result).not.toContain('data-md="-#"');
+    expect(result).toContain('literal caption');
+  });
+
+  it('escapes literal -# when converting paragraph HTML to markdown', () => {
+    expect(htmlToMarkdown('<p>-# plain words</p>')).toContain('\\-#');
+  });
+
   it('converts block math syntax', () => {
     const result = markdownToHtml('$$\\frac{a}{b}$$');
     expect(result).toContain('data-mx-maths');

src/app/plugins/markdown/markdownToHtml.ts

@@ -7,6 +7,7 @@ import {
   maskDollarSignsInsideMarkdownCode,
   shieldDollarRunsForMarked,
   unmaskMathCodeDollarPlaceholders,
+  unmaskSubscriptCodeLinePlaceholders,
 } from './extensions/matrix-math';
 import { matrixSubscriptExtension } from './extensions/matrix-subscript';
 import { matrixEmoticonExtension, preprocessEmoticon } from './extensions/matrix-emoticon';
@@ -150,7 +151,7 @@ export function markdownToHtml(markdown: string): string {
 
   DOMPurify.removeHook('afterSanitizeAttributes');
 
-  const unmasked = unmaskMathCodeDollarPlaceholders(sanitized);
+  const unmasked = unmaskSubscriptCodeLinePlaceholders(unmaskMathCodeDollarPlaceholders(sanitized));
 
   // DOMPurify's Node/JSdom build can drop <img> size attributes even when allowlisted.
   // For Matrix custom emojis, always emit a stable height so outgoing messages have

src/app/plugins/markdown/utils.ts

@@ -78,7 +78,12 @@ export const escapeMarkdownInlineSequences = (text: string): string => {
     (t) => t
   );
 
-  return parts.join('');
+  let out = parts.join('');
+  out = out.replace(
+    /(^|\n)-(# +)/gm,
+    (_, lineStart: string, hashSp: string) => `${lineStart}\\-${hashSp}`
+  );
+  return out;
 };
 
 /**