XML Formatter and Beautifier

The document is tokenised before anything is indented, which is the difference between a formatter and a search-and-replace. A comment, a CDATA section, a DOCTYPE and a processing instruction are each one token, and their contents are never examined — so a "<" inside CDATA is data, not the start of an element, and a formatter built on regular expressions reindents it and quietly corrupts the payload.

Finding the end of a tag is the other place this goes wrong. Searching forward for the next ">" splits a tag whose attribute value legally contains one, such as title="x > y", and everything after it is misparsed. The scan here tracks quoting, so the closing bracket is the one that actually closes the tag. What is inside a tag is then re-emitted verbatim rather than tidied: normalising whitespace between attributes means parsing attribute values, and an attribute value can contain exactly the characters a naive pass would split on.

An element holding nothing but text stays on one line, because a document of leaf values split three ways per element is harder to read than the input was. Structure is checked as it goes and reported rather than thrown: a mismatched closing tag, an element that is never closed and a stray close are each named, and you still get the formatted output, because a half-written document is exactly what somebody is trying to read when they reach for a formatter.