Subject: Re: [netatalk-admins] current netatalk plans
From: wesley.craig@umich.edu
Date: Tue Mar 17 1998 - 01:27:36 EST
> From: bsmith@h-e.com
> To: netatalk-admins@umich.edu
> (An aside for the programmers: I see in the code that FNUMs for files
> are derived from Unix inode numbers, so they are more or less fixed;
> why can't DIDs for folders be derived from the inode numbers of
> directories? Wouldn't that solve the problem, or at least be a
> temporary fix until it can be done "right"?)
The Mac implementation allows us to return FNUMs that are unique only
in the given directory, i.e. the inode number. Since the inode number
of the root of any given mounted directory is 2, it's not practical to
use inode numbers for DIDs -- in /, say, /usr and /var could both be
DID 2.
I believe that a correct implementation must save the DIDs and FNUMs as
part of the Desktop. A system that uses a separate daemon, a la AUFS,
won't work very well in a network file system like NFS or AFS.
However, the system must be robust enough to allow both writable and
read-only version of the database. I think it's mostly just a lot (a
*lot*) of detail work, once it's all spec-ed out.
:wes
This archive was generated by hypermail 2b28 : Sat Dec 18 1999 - 16:31:41 EST