On (10/30/08 17:19), Erik Nordmark wrote: > > sowmini.varadhan at sun.com wrote: > > >> - All commands will take dladm-like flags for persistence, parseable >> output, column selection > > I have no problem with persistence being the default. But I'm concerned > that having temporary as an option for everything might add significant > complexity without solving any real problem. > > In general (thus this might apply to dladm as well) having an > administrator do experiments is a good thing. But having objects that > can have both a temporary and persistent existence, coupled with object > dependencies, makes things difficult both for the implementation (need > to check that a persistent object doesn't depend on a temporary object) > and for the user (need to understand what to do when an error is > generated by the implementation checking the above).
that's a good point. For the features being listed in the current proposal, the only operations that do not have the above dependancy are those that apply to all interfaces (i.e., the setting property with no interface specified). For the rest, one approach is to say that when a specific <interface> is targetted, the persistence attribute of the operation is derived from the persistence attribute of the <interface> itself. Alternately (which is what I think you are suggesting) we could make this simple and say that all commmand that target a specific <interface> are automatically persistently applied. --Sowmini
