[rdiff-backup-users] [PATCH] Fix for bug in regressing

[Josh Nisly]

While doing some unrelated testing, I ran across a bug in regressing a 
failed backup. The regress fails with:

OSError: [Errno 2] No such file or directory: 
'.../run_backups_test/rdiff-backup-data/mirror_metadata.2008-05-08T18:24:21-05:00.snapshot.gz'

It seems that other users have encountered this as well:
http://lists.gnu.org/archive/html/rdiff-backup-users/2007-01/msg00011.html
http://lists.gnu.org/archive/html/rdiff-backup-users/2008-01/msg00033.html

And I believe this bug also started off Sam's problems here:
http://lists.gnu.org/archive/html/rdiff-backup-users/2008-02/msg00018.html

The problem is occurs when we are trying to regress the mirror_metadata 
file, and the file is compressed. The code in recreate_meta() in 
regress.py creates a temporary filename, tells the metadata writer to 
create the file, then moves it to the correct location. The problem is 
that the metadata writer may actually create the filename with a .gz 
extension, so when recreate_meta() tries to rename the file, it doesn't 
exist. (At that point, since the source doesn't exist, it tries to 
delete the destination file, which is why the error message references 
the filename it does.)

I believe the correct solution is to pass a callback into the metadata 
writer that is told what the actual filename written is. Attached is a 
patch to do this.

The fix is pretty simple and obvious, but I do have a (~15MB gzipped) 
repo that demonstrates the problem. I'd prefer not to post it publicly, 
but if a committer wants it to confirm the bug and fix, I can make it 
available to them.

Thanks,
JoshN

rdiff-backup-patch.tar
Description: Unix tar archive