handling renamed files

[dean gaudet]

hi ben,

low priority feature request :)

have you considered a less bandwidth intensive method of handling renamed
files?

i'm seeing lots of churn in /var/log -- because i'm still using a scheme
which renames log.n to log.n+1 each day (i'll probably switch to a dated
scheme instead).

i've also got lots of maildir mailboxes -- which use characters in the
filename to represent status of the messages.  (no quick fix which i can
use here.)

it's a bit tricky to do this...

one scheme which might work would be to have a database on the mirror
which has a list of the source inode numbers for each file in the mirror.
then when a file isn't found at its current name in the mirror, an extra
check is performed to find its old inode ... and if it looks like a rename
occured (i.e. similar file size) then run the rdiff on it.

in the increments directory you could store a newname.T.rename file which
contains the oldname (i think that's the direction you'd need to track the
changes while restoring).

i assume you've already got the inode listing on the master because you're
handling hardlinks now...

-dean