proposed PAG handling changes for Arla

[Lyle Seaman]

The ability to attach to a pre-existing PAG was not a design feature.  It
was an unfortunate side-effect of using groups.  Nor was the use of groups a
design feature, it was merely the most expeditious way of providing the
desired inheritance behavior.  It generated its fair share of problems over
the years.  A more general mechanism (such as a convention that one pointer
in the u area would be available for third-party use) would have been far
better, but required OS support. 

Some tools have been written which take advantage of the ability to swap
PAGs, but they could use another mechanism just as easily, which could have
its own capability.  

Which is not to say that I think restricting setgroups is a *good* idea, I
just don't think that it's necessarily a bad one, either.