I have a bash script that deals with copying very large (>4GB) files from one folder to another. To prevent loss of data I copy the file, check that the file sizes are equal, and then remove the old one. I'm looking for a way to determine when two files in different directories are on the same device so that I can merely rename the file. There most definitely will be cases of cross-device symlinking (shown in the example below) which prevents me from using something like `dirname`. I'll result to parsing a mount listing if there is no simple command(s) to resolve a file to its device.

mounts:
/dev/hda1 -> /
/dev/hda2 -> /mnt/a
/dev/hdb1 -> /mnt/b

symlinks:
/mnt/a -> /mnt/c/a/
/mnt/b -> /mnt/c/b/

files:
/mnt/c/a/old.file
/mnt/c/b/new.file

It would appear the two files are both on /dev/hda1 but in fact the old one is on /dev/hda2 and the new one is on /dev/hdb1. This is just a simple example to show my problem and in no way represents the actual directory structure (which is more complicated).

Use 'stat -c "%d" filename'. It gives you the device id of the file (along with loads of other stuff). The device id's are unique for each disk/partition, so it should serve your purpose.

Thank you very much. I'm glad the solution was so elegant and I didn't have to resort to a ghastly mount/directory interpreter.

You can also use df:

df myfile

I may be misundersanding your problem, sorry if so, but the mv command will rename a file, including moving it from one directory to another, for directories on the same partition/filesystem, and will make a copy if the move is to a different filesystem. It does some error checking and will not erase the original if the copy is unsuccessful. It also has some backup options you may find useful.

The mv command indeed is atomic this way, but it is important to note that neither mv nor the OPs size comparison guarantees identical copies. If fidelity is critical, then perform checksums on the files and compare those as well. Unfortunately, mv makes this impossible, as once the copy (to a new file system) is complete, the original is unlinked.

So, for critical data, where the is the possibility of error, a copy/checksum/unlink operation is best.