Which Markdown parsers apply emphasis delimiters inside words under CommonMark flanking rules

Answer first

CommonMark-compliant reference implementations apply left and right flanking rules to emphasis delimiters. The canonical implementations are the C library cmark and the official ports that implement the CommonMark spec: commonmark.js, python-commonmark, and commonmark-java. These parsers treat * and _ differently under flanking and consistently disallow underscore emphasis inside words.

CommonMark 0.30 and later stabilised the flanking rules. Under those rules underscore emphasis is restricted inside words, while asterisk emphasis is less restrictive and can open/close at word boundaries where the preceding or following character is whitespace or punctuation.

Original Markdown as described by John Gruber does not formally define flanking rules. Parsers derived from the original spec such as Markdown.pl and many early PHP/Python ports historically allowed emphasis inside words for both * and _, leading to divergence from CommonMark. Mixed delimiters like *_text_* are evaluated as separate delimiters with their own flanking checks in CommonMark, which remains a common source of inconsistency in non-compliant parsers.

Confirmed facts

• CommonMark defines left and right flanking for * and _. The reference implementations apply them.
• Underscore is not left-flanking if the preceding character is alphanumeric or punctuation and the following character is alphanumeric, and not right-flanking under the symmetric condition. Hence _foo_bar_ is rendered literally in CommonMark-compliant parsers.
• Asterisk can be flanking at boundaries such as whitespace or punctuation, so *a*b* can be parsed as emphasis in boundary cases while *foo*bar* is not, depending on flanking.
• CommonMark 0.30+ stabilised these rules. Version sensitivity matters for older forks and partial CommonMark modes.

Likely explanation, with uncertainty

Many libraries advertise CommonMark support but deviate on emphasis edge cases, especially for underscores inside identifiers, URLs and code-like text. Mixed delimiter and consecutive delimiter sequences are the most common source of inconsistent rendering even among libraries that claim compliance. Exact behaviour for *_text_* and _*text*_ depends on the flanking status of each * and _ individually, not as a combined token.

Original Markdown leaves treatment of _foo_bar_ and *a*b* implementation-defined. There is no single authoritative behaviour for emphasis inside identifiers across original Markdown implementations.

Steps needed for this case

1. Test underscore inside a word: _foo_bar_. A CommonMark-compliant parser should output literal text without emphasis. A legacy original Markdown parser may render emphasis.
2. Test asterisk boundary cases: *a*b* and a*b*. Compare output for emphasis spans versus literal asterisks to confirm flanking.
3. Test mixed delimiters: *_text_* and _*text*_. In CommonMark each delimiter is evaluated separately under flanking; record whether emphasis is applied and whether nesting is produced.
4. Run the same inputs through the parser you use in production and compare results to the expected CommonMark flanking behaviour for underscore versus asterisk.

To give a definitive recommendation for your documentation, which parser and version are you using in production?