On Mon, 2005-12-05 at 11:22, Joerg Schilling wrote: > Darren J Moffat <[EMAIL PROTECTED]> wrote: > > > On Thu, 2005-12-01 at 11:30, Joerg Schilling wrote: > > > If you take this seriously, then ZFS could not have been allowed to be > > > released > > > the way it has been, because SVN_27 introduces incompatible changes in > > > the ACL > > > interface that would have to be addressed before.... note that these > > > incompatible changes cause problems in star. > > > > I disagree with you here and so did the ARC that this is an incompatible > > change. It is a set of ACLS for NFSv4 and ZFS filesystems with a > > compatible change to the existing system call - ie binaries still worked > > as they did before they just can't backup the new ACLs. This just > > wasn't possible the new ACLs have information in them that you just > > couldn't express with the old ones; plus they are what customers want > > and need and backup and archiver software will just have to change or > > become irrelevant. > > If you run > > star -c -dump -acl > > on a ZFS tree you should find out your self that there indeed was an > incompatible change...
Thats where we disagree. Does this change cause star to break on UFS filesystems ? Not that I'm aware of. It means that star doesn't know about the NEW ACLs but doesn't break the old ones. This means that star needs to be updated to deal with the NEW ACLs. That IS a compatible change - it might not be the change you would have liked but that doesn't make it incompatible. The compatibility guarantee that we make with Solaris requires that old code and binaries continue to work if they use documented and Stable interfaces. We didn't break that promise with the addition of a NEW file system with a NEW ACL system even though we used the already existing system call. > well, you could call it a bug. I wouldn't I'd call it a request for enhancement in star, providing we didn't break star -c dump -acl for archive creation and expansion on UFS (not transition between UFS and ZFS in either direction though). > But why does acl(info->f_name, GETACLCNT, 0, NULL) sometimes return > an error code that is not listed in the Solaris 10 man pages > with either code or reason? Look at the source to find out why. If there is an errno returned that isn't documented on the most recent man pages then please log a bug on that. -- Darren J Moffat _______________________________________________ opensolaris-discuss mailing list [email protected]
