fix: keep a backslash in a code span from hiding emphasis

[breken-ai]

Marked version: 18.0.14 (master at 4ee44f7)

Markdown flavor: CommonMark (also GitHub Flavored Markdown)

Description

No existing issue that I could find.

Expectation

A code span that ends in a backslash does not affect emphasis later in the paragraph. This line from the minimatch README:

Note that `\` or `/` _will_ be interpreted as path separators in paths on
Windows, and will match against `/` in glob expressions.

should render <em>will</em>, as it does in the CommonMark reference implementation, markdown-it and on GitHub.

Result

<p>Note that <code>\</code> or <code>/</code> _will_ be interpreted as path separators in paths on
Windows, and will match against <code>/</code> in glob expressions.</p>

The underscores come through as literal text. The same happens with **bold** in the glob README (`\` characters to `/` in pattern strings, **thus making it impossible...**) and with any Windows path in code, e.g. Use `C:\Users\` for **home** and `D:\` for data. Minimal case: `a\` *b* `c`.

What was attempted

In #inlineTokensInner, the em/strong mask replaces every backslash escape (anyPunctuation) before blockSkip masks code spans. A backslash is literal inside a code span, but the mask still turned \` into ++. That removed the closing backtick of `\`, so blockSkip paired the opening backtick with the next code span's backtick and masked everything in between, including the emphasis delimiters. This ordering dates from #2838.

The fix adds anyPunctuationOutsideCode. It matches a code span (same shape as inline.code) or an unmatched backtick run whole and leaves it unchanged, and masks only the escapes outside code spans. Escaped backticks outside code still work: \`*escaped tick*\` keeps its emphasis. anyPunctuation itself is unchanged because the tokenizer also uses it to unescape hrefs, titles and info strings.

Test: test/specs/new/codespan_backslash_emphasis.md. It fails on master and passes with the fix. npm run test:specs (1865/1865), test:unit, test:cjs, test:umd, test:types and test:lint all pass. As a further check, I compared this branch with master on 300k generated inputs mixing backticks, backslashes and delimiters. Every output that changed now matches commonmark.js (or markdown-it with gfm: true), and none moved away from it.

This change was prepared with the help of an AI coding assistant, and I reviewed and tested it.

Contributor

• Test(s) exist to ensure functionality and minimize regression (if no tests added, list tests covering this PR); or,
• no tests required for this PR.
• If submitting new feature, it has been documented in the appropriate places.

Committer

In most cases, this should be a different person than the contributor.

• CI is green (no forced merge required).
• Squash and Merge PR following conventional commit guidelines.

[vercel]

@breken-ai is attempting to deploy a commit to the MarkedJS Team on Vercel.

A member of the Team first needs to authorize it.

[vercel]

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

Project	Deployment	Actions	Updated
marked-website	Ready	Preview	Oct 8, 2026 2:44am UTC

[UziTech]

Thanks! 💯

[styfle]

Looks good though we deviate a bit from commonmark. Here's an example:

https://spec.commonmark.org/dingus/?text=_%60%60%5C%60_%60%60

[breken-ai]

@styfle thanks for the example. It was a regression from this PR. master renders _``\`_`` as <p>_<code>\`_</code></p> (the same as the CommonMark dingus), but the earlier head 521ec76 rendered <p><em>```</em>``</p>.

Here is why. Before this PR, masking \` as ++ happened to leave that code span with no backticks inside, so blockSkip ((?<b>`+)[^`]+\k<b>(?!`)) masked the whole span. Once the PR stopped masking escapes inside code spans, the inner backtick stayed, blockSkip no longer matched the span, and the _ inside it closed the emphasis.

Fixed in 7d5c8c9. Inside a code span that the escape mask matches, the inner backticks are now masked to + (the length stays the same), so blockSkip skips the whole span again. The other characters are untouched. I added two cases to test/specs/new/codespan_backslash_emphasis.md:

_``\`_``

_`` a`_ ``

These now render as <p>_<code>\`_</code></p> and <p>_<code>a`_</code></p>. On this spec file the fix matches the commonmark reference CLI byte for byte. The spec fails on 521ec76 (Actual: …<p><em>```</em>``</p>) and passes on 7d5c8c9. test:specs passes 1865/1865, test:unit passes 193/193, and test:lint and test:types are clean.

[UziTech]

Thanks for the fix!