[bug-diffutils] bug: diff -r swaps the files it should check

[Christoph Anton Mitterer]

Hi.

I found a rather strange bug in diff (Debian version 1:3.0-1) when doing
large backups of millions of files.
First I've copied the data to an external HDD, then I've did an "diff -q
-r >out 2> err" in order to find any errors.

out has about the following content:
Files /data/......../WeiᳮJPEG and
/mnt/backups/2010-08-06/1/......../WeiᲮJPEG differ
Files /data/......../WeiᲮJPEG and
/mnt/backups/2010-08-06/1/......../WeiᳮJPEG differ

The strange unicode characters had their character codes inside which read
as follows:
1CEE      1CAE
1CAE      1CEE
(in the same order as above)


stat on these files gave:
$ stat Wei*
  File: `Wei\341\262\256JPEG'
  Size: 20927           Blocks: 48         IO Block: 4096   regular file
Device: 821h/2081d      Inode: 4096614     Links: 1
Access: (0660/-rw-rw----)  Uid: ( 1000/calestyo)   Gid: ( 1000/calestyo)
Access: 2009-11-30 19:29:18.000000000 +0100
Modify: 2001-07-18 16:40:08.000000000 +0200
Change: 2007-07-19 19:12:28.000000000 +0200
  File: `Wei\341\263\256JPEG'
  Size: 15524           Blocks: 32         IO Block: 4096   regular file
Device: 821h/2081d      Inode: 4096615     Links: 1
Access: (0660/-rw-rw----)  Uid: ( 1000/calestyo)   Gid: ( 1000/calestyo)
Access: 2009-11-30 19:29:18.000000000 +0100
Modify: 2001-07-18 16:40:08.000000000 +0200
Change: 2007-07-19 19:12:28.000000000 +0200

There was even a third file of similar name:
  File: `WeiᮊPEG'
  Size: 18843           Blocks: 40         IO Block: 4096   regular file
Device: 821h/2081d      Inode: 4096613     Links: 1
Access: (0660/-rw-rw----)  Uid: ( 1000/calestyo)   Gid: ( 1000/calestyo)
Access: 2009-11-30 19:29:18.000000000 +0100
Modify: 2001-07-18 16:40:08.000000000 +0200
Change: 2007-07-19 19:12:28.000000000 +0200
but it seems that this one made no problems?


When manually comparing those files with diff (of course in the correct
order),... no difference was found.


Any ideas?



Cheers,
Chris.

btw: I'm not subscribed, so please CC me.