For example emitters need to know what void elements are because <br></br> is actually equivalent to`<br><br>`. But `<script src=foo.js/>` is only an opening tag so the rest of your document will be executed as JavaScript. So you can't just write an emitter for arbitrary elements, you need to emit different things for `br` and `script`. Plus `script` has special escaping rules that are often forgotten about. Plus you better keep that list up to date!
With XHTML it is very easy to write a parser that will construct a tree forever and can reserialize it with no issues. I have no issues with consistent changes such as empty attributes and unquoted attribute values, but I think that these element insertion, auto-closing, void elements and non-replaceable character data are a mistake because you need to maintain an up-to-date dataset of these custom rules or you get an incorrect result.