Before I get started with my $.02, let me point out that although I
can influence decisions regarding new features, that this message
should not be misconstrued as a indication to do so or as a guarantee
of groups within groups in a future release. \*end disclaimer*\

My concern regarding groups of groups has been loss of control of the
big picture by the person who owns a group that contains someone elses
group.  For example, lets say I have a group

%  pts mem  pierette:co-workers
    phil
    cindy
    linda:co-workers
         kt
         mary
         alan
         kt:people
             keith
             jeff:friends
                  etc
                  etc
                                    
By putting linda:co-workers into my group - I effectively gave some
administrative control of my group to Linda, who *later* added the
group kt:people to her own group ...... effectively giving some
adminstrative control of my group to Katie.  Linda is probably unaware
of the fact that by modifying her group, she's also modified mine -
and she really shouldn't *have* to worry about it. Jeff probably has
no idea that his friends are becomming my co-workers and I'm likely
unaware of this too.

So if used by "normal" users, groups within groups may quickly become
a real quagmire.  How do you adequately warn naive users of this phenomina?
I'm not convinced they'd think of it themselves (after all, none of
the experienced users on this list mentioned it.)

One may argue that the ability to place groups in groups should be
limited to system:administrators, but on the scale that UMich is
discussing, I think that would be an unacceptable limitation.

Consider this for a solution.

For non system:adminstrators, groups can only contain groups that are
owned by the same "person".  Ie, I can have this setup

% pts mem pierette:people
        pierette:managers
                linda
                kt
        pierette:co-workers
                pierette:training
                pierette:documentation
                pierette:afs-developers
        pierette:friends


More importantly, the History department at UMich can have

% pts mem history:folks
        history:faculty
                sue
                chan   etc
        history:students
                history:undergrads
                        bob
                        nancy   etc
                history:grad-students
                        george
                        mcintyre  etc
        
And, the sys admins at Umich can put any group into any group, so
knowing that the only members of the prefixless group "history" are
the department head and their administrative assistant, they feel
confident that all the history groups (and math groups, and chem-e
groups, etc) are well managed.  So they (the sys:admins) make groups
like this

% pts mem Umich:people
        humanities
                history:folks
                english:folks
                etc
        engineering
                chem-e:folks
                ee:folks 
                etc

% pts mem Umich:faculty
        history:faculty
        english:faculty
        ee:faculty
        etc
                      
Now I haven't even begun to look at the code to see how feasible this
is.  And there are certainly things to be considered, such as the
effect of a

% pts chown pierette:friends -owner david

and how that would effect the group pierette:people

There is clearly a strong desire for groups within groups, and at this
stage many of you may be so enthusiastic about the benefits that you
may be willing to overlook the "loss of control" problem that my
afflict your naive users.  Maybe it's not nearly the problem that I
fear it is.  Nonetheless, I'm out there fighting for the little guy,
Joe Freshman, who finds out through some weird twist of groups w/i
groups that his worst enemy is reading all his personal files because

        joef:best-friends

is more that he ever expected or planned it to be.  

Now imagine this happening to Joe T. Professor (T is for tenured!).

Pierette VanRyzin
AFS Training

        






Reply via email to