I've fixed all known xml2po bugs (at least all bugs reported in Bugzilla; I know of at least one other bug [look at TODO] which I only partially fixed :).
But, this would be no good if I didn't actually start developing regression tests. I'm not planning on distributing them, but I've put them in CVS so anyone can look at them and try them.
Since it's sometimes hard to come up with decent examples, I'd like to have a couple of more complex XML (preferably DocBook) files, which will push xml2po to it's limits. Of course, it's better if they're smaller, but I won't mind even large, very complex documents, but only if you provide me with a translated PO file as well (so I can test both extraction and merging).
Now, off to cleaning the bathroom. :)