[gnodet]

Problem

The chhorz javadoc-parser wraps {@code X} content verbatim in <code>X</code> without HTML-escaping X. When X contains angle brackets (e.g. {@code <element>}), Jsoup subsequently parses <element> as an unknown HTML tag and silently drops it from the XML output — the tag name disappears from the generated site.

This is the root cause of apache/maven#13675, where descriptions such as:

• {@code <activeProfiles>}
• {@code <activation>}
• {@code <name>value</name>}

all render with the angle-bracket content missing on the published Maven site.

Root cause trace

In XdocGenerator.rewrite():

1. The MXParser XML parser reads a <description> outside CDATA and decodes < → <. So {@code <element>} in the .mdo file becomes {@code <element>} in the description string.
2. The javadoc-parser produces <code><element></code> — the content is not HTML-escaped.
3. Jsoup, in XML output mode, treats <element> as an unknown element and drops it silently.

Fix

Pre-process {@code} and {@literal} inline tags in XdocGenerator.rewrite() to HTML-escape their content before the javadoc-parser and Jsoup pipeline see it.

The escapeHtml() helper normalises pre-existing </> entities first (which appear in CDATA-wrapped .mdo descriptions) so both forms are handled correctly and produce identical correct output:

• {@code <element>} (XML-decoded) → <code>&lt;element&gt;</code> ✓
• {@code <element>} (from CDATA) → <code>&lt;element&gt;</code> ✓

Test

Added a field with {@code <name>value</name>} (outside CDATA, so MXParser decodes it to {@code <name>value</name>}) to html4.mdo and verified the expected output <code>&lt;name&gt;value&lt;/name&gt;</code> in html4.expected.xml.

[gnodet-bot]

preprocessInlineCodeTags replaces both {@code ...} and {@literal ...} with <code>...</code>, but the chhorz parser maps {@literal} to <i>...</i> (italic), not <code>. By preprocessing both tags to <code>, the method silently changes the rendering semantic for @literal.

Two smaller issues noted as well.

This review was generated by an AI agent, Hermès on behalf of @gnodet.

[gnodet-bot]

Following up on my earlier review — all 3 findings have been addressed.

✅ {@literal} rendering — fixed by capturing the tag name via m.group(1) and using the ternary to emit <i>...</i> or <code>...</code> accordingly.
✅ Nesting comment — qualified to "one level of nested braces".
✅ Missing trailing newline — added to html4.expected.xml.

Pre-fix validation: tested XdocGeneratorTest#testHtmlToXml against the base commit — test fails (Jsoup silently drops <name>value</name> leaving <code></code>). With the fix it passes. The test correctly exercises the fixed code path.

The implementation is clean: escapeHtml() normalises pre-existing entities first to avoid double-escaping, and the regex correctly limits nesting to one level (as now documented). LGTM.

This review was generated by an AI agent, Hermès on behalf of @gnodet.