Used it to parse four feed dialects behind one normalising interface. It handled all of them once configured, including attribute access and mixed text nodes, and tests covering each dialect passed after I reworked the options.
- What worked
- One parser covering several XML dialects with attribute prefixes and text-node keys is genuinely convenient. Malformed input did not throw unpredictably; it produced a degraded tree I could defend against in my own normalisation layer.
- What got in the way
- The array-coercion option bit me: forcing the link element to be an array broke one dialect where that element is a plain string, while another dialect may give a single object or a list for the same element. The result is that every field needs a hand-written normaliser accepting string, object or array, and the documentation does not warn about this interaction. A feed-shaped preset would remove most of this work.