[comp.protocols.appletalk] CAP aufs quirk

jmm@cello.mc.duke.edu (James Madill) (04/10/90)

In article <51927@coherent.coherent.com> dplatt@coherent.com (Dave Platt) writes:
>On Mac hard disks, each directory is assigned a unique
>identifier when the volume is created;  the ID for a directory is never
>changed, and the IDs are never re-used.

>Aufs does not... it dynamically creates an ID for each directory-path
>which you access, uses this ID for the duration of the Aufs session, and
>then (when the Aufs process shuts down) discards the IDs.  There's no
>guarantee that the same ID will be assigned to the directory during your
>next session.


I had figured this out from the documentation but the odd part was that I
didn't quit my aufs session; I didn't even exit Excel.  I created the link,
closed the dependent file, changed one of the numbers in the calculation,
closed the independent file, and then reopened the dependent file.


>If you're in the mood for it, you might be able to hack Aufs to ensure
>constant directory-IDs across multiple sessions.  You might, for
>example, be able to rewrite the code in applications/aufs/afpdid.c to
>use the directory's inode number as the basis for the external directory
>ID.  This would work, I think, unless your Aufs directory-hierarchy
>contains symlinks, and up until the moment at which you start copying
>directories around or backing them up and restoring them.

I've been considering using the inode number as the directory id.  I
believe the number assigned to the directory is static throughout its life.
If it is deleted, then the data within that directory is also deleted and
thus the link is non-existant.

Since excel 2.2 has been around for awhile, I was hoping that someone had
already made this hack/patch/rewrite.  The department needs an administrative
file server that speaks AFP and the decision on what hardware to purchase
can be delayed somewhat but we are getting a new department head in July and
we would like this up, running, and debugged by the time he arrives.




James Madill				Internet: jmm@cellbio.duke.edu
Box 3709 DUMC				Bitnet:   madil001@dukemc
Duke University Medical Center		work: (919)684-3049
Durham, NC 27710