* James Carlson <james.d.carlson at Sun.COM> [2007-05-01 09:05]: > Dermot McCluskey writes: > > That's exactly what I started doing. But after over 100 lines of > > awk, I realised there were lots of scenarios I still wasn't taking > > care of in the install CAS, and I hadn't even started looking at > > the remove script yet. So yes, it is reasonably complicated. > > I see. > > > I think SMF provides a much more suitable way to update > > these files. Is there a general policy or desire to avoid > > proliferation of SMF services? > > It's a bit icky. It's somewhat akin to self-modifying behavior, in > that the installed software goes through some sort of lengthy > "rebuild" after the first boot. > > There are a number of issues with that, including how it works with > diskless configurations (it probably just doesn't) and what happens > with packages that can be added and removed on the fly. > > We have other examples of this (such as fc-cache), so I'd say it's not > impossible, but I'm also not sure that it's something I'd like to see > become commonplace. (Perhaps fc-cache, with its frequent core drops > and scary upgrade behavior is an extreme example ...)
I think that restricting these tasks to scenarios when the system is running with a known filesystem root is less complex than our current approach, where each class action or packaging support script needs to correct its operations for liveness and alternate root-edness. I can't speak for the fc-cache case specifically. - Stephen -- sch at sun.com http://blogs.sun.com/sch/
