Run Swingset. Switch to the JTable test (next to last tab). Position some
overlay window (e.g., curPal, which you can select to stay on top) over
the scrolling area so that it creates a complex (non-rectangular) clipping
region in that areay. Pull the scrollbar up and down. Notice how the
repaints do not happen correctly and lots of garbage is created. For example,
note how pars of the obscuring window are copied into the contents of the
JTable area.

Sometimes this is unreproducible, but I was generally able to reproduce it
by placing the overlay window over the JTable area before the first scrolling
action. If it did not reproduce that time, I would move the overlay away,
scroll, and then move the overlay back; it usually showed up the problem
after that.