That can happen on the 2nd use - if 1st time it was closed, and this way returned to the pool, it'll still closed when it's given again to application. Actually, need to verify if it was really in 1.1.2, or regression brought after the last release

Still not quite clear - on one hand it seems like it should be covered by old testcase(s), on other - I can't see in changes after last release what could cause it. I am inclined to leave it in place.

If return to the pool was initiated by close(), it was possible, that new connection already got the returned connection while close() still continued routine, including marking underlying Protocol object as closed.
Plus there were other cases possible - it's described in the commit description