On Wed, 04 Aug 1999 08:54:39 +1000  Giles Lean wrote:

[Following up my own posting ... sigh]

> I can match the garbage files to an original that has the same length
> and is correctly readable.  (I've added logging so that pathnames will
> be recorded too, just in case.)

I've noticed that the process doing the copying is correctly reading
the original file -- it makes correct choices about where to file the
copy.

Now, is perl's memory being scrambled, is there a write error, or is
there an error on reading the data back from the filesystem.  Reading
the data off the raw device indicates that the garbage is definitely
on the disk, so the buffer cache is innocent.

Giles