[ILUG] Hard link copying peculiarity

[Niall O Broin]

I had 43GB of JPEGs to copy from one disk to another. Each file has two
directory entries i.e. two hardlinks which I want preserved (because I didn't
want to copy 86GB of data to the destination disk). For bulk copying of
data, I tend to do

find * | cpio -pmd destdir

Another option is to use cp -a (I'm using find | cpio for years, something I
started to do on Unices which didn't have a -a option to cp) and another is

tar cf - * | (cd destdir; tar xfBp -)

which I used to use before I started using find | cpio.

However, both the cp and cpio methods failed for me here - they didn't
maintain the symlinks and filled the destination disk. What's curious is
that I knew this was an issue and I did a test run without about 1GB of data
and that worked fine - the files were copied c/w hard links.

I was about to try using the tar method when Paul Kelly suggested using
rsync (a great program, and one I use a lot for syncing remote and local
files, and also for syncing two local disks - but I don't tend to think of
it for simple copying) and a quick perusal of the manual showed that it had
a specific command line option (-H) to preserve hardlinks. 

To cut a long story short, rsync did the business. The whole thing however
leaves a sour taste in my mouth WRT cp and cpio though. I find it very
strange and disturbing that these programs will work in two different ways
depending on how much data is being copied.

OK - just had a quick look at the cpio source - it maintains a hash of
filenames and inode numbers. I'm no algorithm expert but I suspect that
perhaps this hash implementation falls over when presented with ~100000
entries.




Niall