On Sat, 17 Jan 2009, Andrew Ferguson wrote:
The reason it hasn't been implemented is that it would require
either figuring out how much space is needed beforehand (which
requires scanning the whole source repository in advance, something
which rdiff-backup is currently not setup for; however implementing
that functionality would allow various other requested features to
be developed), or rdiff-backup would have to detect that out-of-
disk-space event, reverse the current session, delete an increment,
and start over (that, of course, hits a horrible case when the
current backup wants to add a, say, 40GB file, and deleting each
increment only frees a few MB or so).
Surely it's possible to delete an old increment without reversing
the entire backup in progress? As I understand it, a backup creates
a new increment, it doesn't touch any of the existing ones,
therefore it should be perfectly safe to delete any increment except
the one currently being created, while the backup is running. At
worst, we'd have to restart the backup of the current file, if the
out of space error left the binary diff
file hopelessly corrupted.