I have discovered that:
* This happens every time I give it a disk which has the layout I
described in my first report.
* The presence of cciss block devices is a red herring; I have
reproduced the problem on a machine without them. Indeed
the hardware doesn't appear to matter.
* Whatever the partitioner does before it crashes destroys the
preconditions for the bug.
I have captured an image of a 400G hard disk in the bug-triggering
state. After bzip2 it's only 2.1Gby so I have put it on a USB stick
and will hope to give it to Colin Watson in person. md5sum:
I can't publish the image because it contains a copy of Windows.