When we moved the logic to handle the first one, wrong loop logic was
kept in place which meant we still finished early. But we now notice it
because we're not reading past the last LF we find.

This was not noticed before as the last field in the tested commit was
multi-line which does not trigger the early break.