Thanks for the explanation. I think we should revert it. IMHO it was
a nice clean example of a streaming transformation, but unfortunately
it transformed an API that nobody liked in the first place, and broke
some weird and wonderful workarounds. Let's try again in 18.