Having a background in C/C++, that was the problem I ran into when I had to learn XSLT at translation company that used it to style documents across multiple formats. The upside of using XML was that you could store semantically rich info into the tags for the translators and designers. The downside, of course, with all the metadata, was that the files could be really large and the XSLT was usually specifically programmed for that particular document and very verbose so the XSLT template might only be used a couple times.

XSLT is really strange in that it's not really what people think it is. It's really a pattern-matching and production rules system right out of the golden age of AI but people think it is just an overcomplicated Jinja2 or handlebars.