Another idea

[Tobias Polzin]

>   TP> The best way would be, if rdiff-backup supported the "cp -l" way
>   TP> of copying: Create hard-links instead of copying. Do you think
>   TP> this would be possible to implement.  I assume, that I just have
>   TP> to add another switch and update Robust.copy and
>   TP> Robust.copy_with_attributes.  What do you thing?
> 
> Do you mean that when you run something like "rdiff-backup indir
> outdir" that the stuff in indir gets linked inside outdir?  I think
> you are right about it maybe not being difficult to implement, but I
> don't see the utility of a switch.  Most of the time indir and outdir
> would not be on the same filesystem so hard linking isn't an option.
> Also it makes it easy for users to corrupt their backup archive - if
> indir is hardlinked to outdir and someone edits a file in indir it
> will cause outdir to change too, and the archive gets corrupted.

It is useful for all those systems, where you don't have python 2.2
and librsync on your local machines, just a way to bring data to a
common backupserver, that has everything needed installed.

I don't think that it's only the case for me...

But I agree that it is a security problem. The possibility to ruin your
backup is always a problem. In the link you send me, this problem is
solved in an elegant way.

As it seems, that you don't like the idea, I will have to implement it
myself. Shall I send you patches that you have it (at least as undocumented
feature) and I can more easily incooperate rdiff-backup updates?

> Shouldn't be too much of a problem - I just always seem to end up with
> 97% of the space used...

So, what fraction "mirror files"/"increment files"?

Tobias