Re: [rdiff-backup-users] Are my backups too corrupted to recover?

[JoSH Lehan]

Andrew Ferguson wrote:
> Hmm, interesting. I was just thinking about this bug the other day and 
> hoping to figure it out.

That would be great if the "Bad index order" bug could be fixed.  It 
stings many people when trying to regress.

Some data points I have noticed:

1) It only happens for filenames that had to be remapped to numbers 
within the "long filename data" directory.

2) It only happens during rollback of a previous backup run that failed 
somehow.

3) This bug will often happen for filenames with Unicode characters in 
them.  How exactly does Python assign a sort order to tuples?  I'm 
surprised the rdiff-backup code would even care about the order of 
filenames at that point.  I"m hoping that there's not some kind of 
locale screwup that causes Unicode characters to get a strange sort 
order that is different from what Python expects.

I haven't had this bug happen in a while, but the last time it happened, 
I tried running the regress both in the "C" locale and the default 
(en_US.UTF-8) locale, and it still failed for both times, so the system 
locale setting wasn't the problem.  Since then, I haven't been able to 
successfully reproduce the bug.

Josh