> Plus, some major XML libraries, like lxml in Python, are extremely unintuitive in their implementation of DOM structures and manipulation.

Lxml, or more specifically its inspiration ElementTree is specifically not a (W3C) DOM or dom-style API. It was designed for what it called “data-style” XML documents where elements would hold either text or sub-elements but not both, which is why mixed-content interactions are a chore (lxml augments the API by adding more traversal axis but elementtree does not even have that, it’s a literal tree of elements). effbot.org used to have a page explaining its simplified infoset before Fredrik passed and registration lapsed, it can be accessed through archive.org.

That means lxml is, by design, not the right tool to interact with mixed-content documents. But of course the issue is there isn’t really a right tool for that, as to my knowledge nobody has bothered building a fast DOM-style library for Python.

If you approach lxml as what ElementTree was designed as it’s very intuitive: an element is a sequence of sub-elements, with a mapping of attributes. It’s a very straightforward model and works great for data documents, as well as fits great within the langage. But of course that breaks down for mixed content documents as your text nodes get relegated to `tail` attributes (and ElementTree straight up discards comments and PIs, though lxml reverted that).