Hmm, the source certainly looks identical and the test passes for me too... Strange.
The most obvious difference is the database and dialect being used. I encountered the problem with Oracle, the test uses H2.

I will play around with this test a little more on Monday evening.