This probably needs to go upstream… not sure if I already did something about it.
(The document separator definitely needs lots of testing to be upstreamable. We could also try to use a parser that can handle the somewhat "illegally" formatted output.)