[slachiewicz]

MarkdownSink.unknown() logged a warning and silently dropped any Sink event it had no dedicated handling for, taking the content it carried with it (e.g. <object> embeds, image maps). It now delegates tag reconstruction to Xhtml5BaseSink.unknown() and emits the raw HTML tag instead, restoring that content in the converted output. <thead>/<tbody>/<tfoot> stay excluded, since they carry no content of their own and Doxia's table sink API already captures the structure they wrap.

Aligned with #1087 (xdoc-selfclosing-raw-html): a self-closing, non-void tag (e.g. <object/>) is never rendered self-closed, since HTML5 ignores the trailing slash on non-void elements and would otherwise swallow all following content as fallback children. It is written as an explicitly closed pair instead.

Root-cause fix for apache/maven-doxia-converter#159.

Verified: end-to-end conversion of the issue's xdoc fixture (<object type="image/svg+xml" data="x.svg"/>) through a locally built doxia-converter now yields <object type="image/svg+xml" data="x.svg"></object> in the Markdown output instead of silently vanishing.

This change was created with AI assistance.

[michael-o]

Is the behavior of #unknown() defined in the API or left to the implementor, read undefined/impl specific?

[slachiewicz]

Left to the implementor, explicitly. Sink#unknown says:

Adds an unknown event. This may be used by parsers to notify a general Sink about an event
that doesn't fit into any event defined by the Sink API. Depending on the parameters, a Sink
may decide whether or not to process the event, emit it as raw text, as a comment, log it, etc.

"emit it as raw text, as a comment, log it" is an explicit grant of latitude, and the existing
implementations already use all of it: SinkAdapter no-ops, AptSink logs and ignores,
Xhtml5BaseSink reconstructs the tag. So moving MarkdownSink from log-and-drop to emitting the
tag is a choice the contract allows rather than one it constrains.

MarkdownSink extends Xhtml5BaseSink, so this PR delegates to that implementation rather than
inventing its own interpretation, and only diverges where HTML5 forces it to (a self-closed
non-void tag).

This comment was created with AI assistance.