[slachiewicz]

Closes self-closing non-void HTML tags (e.g. <object .../>) before Jsoup re-serializes flexmark's raw-HTML passthrough, so they no longer swallow the content that follows as unclosed-tag fallback children.

Fixes #1086

Verified: mvn test on doxia-module-markdown → all green (new test fails without the fix, passes with it).

This change was created with AI assistance.

[michael-o]

Does this cover the case if the tag spans multiple lines? Which can happen...

[slachiewicz]

Yes. Verified by running the pattern rather than reading it:

input	result
<object type="image/svg+xml" data="x.svg"/>	closed
<object + newline + type="…" + newline + data="x.svg"/>	closed
<object data="a + newline + b"/> (newline inside a quoted value)	closed
<br/>, and <img + newlines + src="a.png"/>	left alone, correctly
<object data="a"/><object data="b"/>	both closed

Multi-line works because the pattern contains no ., so DOTALL never comes into it. The
attribute group is built from "[^"]*", '[^']*' and [^"'/>], and in Java all three match a
newline, as does the \s* before />.

Code blocks are safe for a separate reason worth recording: flexmark escapes their content, so
<object/> never matches and a literal example in a fenced block is not rewritten.

Testing it did turn up one case it misses, though — an unquoted attribute value containing a
slash:

<object data=x/y.svg/>   ->   unchanged

/ is excluded from every branch of the attribute group, so the match can never reach the
trailing />. HTML permits unquoted attribute values and flexmark passes raw HTML through
verbatim, so that input still hits the original bug. Narrower than the multi-line case you asked
about, but it is the real hole. Happy to either handle it or record it as a known limitation with
a test — your call which.

This comment was created with AI assistance.