> > I never suggested we should try to design to handle this case; I am merely > > refuting the claim that folks should "fix the resolver". However, it > > would seem reasonable to me to have ipadm bail out with an error if it > > finds a given hostname maps to more than one IP address. > > right.. another thing to remember here is that the debate about > generating a default user-friendly label for an address is more of an > issue for the CLI than for the library. Backing up a bit, it's only > at the CLI that we have issues like "what is a user-friendly way > to specify an iptunnel to 'ipadm delete-address' from the command > line?"
Agreed. > In the libipadm invocation, one could just call ipadm_create_addr() > (or some lookup function to find an address) and get back an opaque > structure that would then be used to do other operartions on the > address. Right. > So does the following > ipadm create-addr -i <intf> [-a laddr[,raddr]] <label> > offer potential for confusion? > (an address pair will be needed for tunnels. <label> will be > used to generate the local address, when no -a arg is supplied) So the proposal is to use the label to do a hostname lookup and attempt to derive the address from that? Doesn't that only further the strange duality between labels and hostnames? That is, when the admin first does a create-addr, the label doubles as a hostname in that it is looked up to determine an IP address. However, after that point the label is just treated as a label, and even if the hostname is changed to refer to a different IP address, the label (with the same name as the hostname) continues to refer to the old address -- until a delete-addr is done and another create-addr is done with the label, in which case it then gets the latest address. Or have I misunderstood what you're proposing? -- meem
