CSCdi75464, integrated in 11.1(7)CA and 11.1(7.3) mainline.
This is an optimization, not a bug fix, although I'm sure skeptics would subjectively argue otherwise.
How can you claim that this isn't a bug?
Because the protocol spec doesn't say you can't, or even shouldn't, propagate a withdrawal to a neighbor you havn't advertised something to. Conforming BGP receiver implementations issue no warnings and do not behave badly in the presence of extra withdrawals. The total BGP byte count due to extra withdrawals was never very high. The total CPU ticks used to ignore them was never very high. Extra withdrawals was something somebody noticed because they were looking for something else. It was amusing, not alarming. Cisco fixed it because people (like me) thought it was sloppy and just would not shut up about it. Tempest. Teapot. Have none of you got anything more important to discuss?