OS X 10.8.4, 7u40b31, dual-screen setup, retina emulator on a primary device.
Run an old as hills regression test
on OS X. It intersects two non-intersecting rectangles and applies this intersection as a clip. The following getClipBounds() elsewhere returns the same rectangle with negative width and height. This rectangle (as I understand), first, is kept as a clip set; second, is ignored as such (no attempt to apply or repair)(am I right?). Why it is so, see in https://jbs.oracle.com/bugs/browse/JDK-4147957.
Oh, I forgot to say, the test fails on Mac.
The test does pass on my single-monitor Linux and on dual-monitor Windows.
I don't know if retina (emulator) is essential for failure on primary screen.
I don't know if dual-screen setup is essential.