Quoth James Carlson on Thu, Jan 22, 2009 at 05:11:10PM -0500: > > I expect it to age well because there are always users behind > > modifications, and so each modification should be able to be classified > > into a customization or a default. > > Is generic.xml a "customization" or a "default"? How about > platform.xml? Or an applied profile?
generic.xml and platform.xml are defaults because they are written by service authors. A profile applied by "svccfg apply" is a customization because it was invoked by an end-user. > With the current proposal, it sounds like some of these bundles of > bits are treated as part of the system, and thus not a > "customization," while other parts (such as applied profiles) are > treated as customization. > > I think that's a bit arbitrary and potentially very confusing, > especially since one of the (many?) initial users of these things will > be NWAM, which will almost certainly be playing around with > automatically applied profiles. Those things aren't "customizations," > but rather states of the system (or the effects of being in certain > states). If they were specified by an end-user, then they are customizations. If they were specified by a service author, they are defaults. I suspect you're talking about the ability to have different defaults or customizations in different contexts. I don't think that changes whether they're customizations or defaults, though. > > I think it's possible that the packaging system could tell us when an > > upgrade is complete and leftover customizations are probably garbage. > > I think it's also possible that backups become so easily accessible that > > keeping the customizations are more trouble than it's worth. Absent > > either, though, I think it's best to keep them and provide easy ways for > > the administrator to find and remove them. > > That seems like the best for now. > > One possibly interesting result is that an inexperienced user who > uninstalls the software and then reintalls it (hoping, perhaps, to get > a blank slate to "try again") will be surprised. Right, and if we can get the packaging system to tell us whether a manifest is being removed due to a package upgrade or a package removal, we might be able to make better decisions. David
