> I'd be more than happy to drop support for DNS/hostname resolution > and just require numeric addresses to be specified to ipadm. But I gather > (both from Jim and some other users like Clusters) that having the > /etc/hosts resolution with ifconfig was somewhat attractive. > > here's another option: make the "-a" optional, and the "label" always > mandatory. So > > # ipadm create-addr -i net0 myhost > > will lookup myhost in DNS and configure it on net0. It will fail if > myhost resolves to more than one address. If you don't like that, fix > the resolver.
"Fix" implies it's broken. It's standard operating procedure at many sites. > # ipadm create-addr -i net0 -a 1.2.3.4 addr1 > > will set up 1.2.3.4 on net0 and let you refer to the address as addr1 in > subsequent commands. > > # ipadm create-addr -i net0 -a 1.2.3.4 > and > # ipadm create-addr -i ip.tun0 -a 1.2.3.4,1.2.3.5 > > will fail. Please specify a label. > > # ipadm create-addr -i ip.tun0 -a 1.2.3.4 tunaddr > > will fail- 2 addresses need to be specified. > > Is that sufficently confusion free to all? > > > used to, but has a naming mechanism that is distinct from hostnames (but > > Well, logical interfaces are doing that already. No: logical interfaces identify address slots rather than name addresses. This is why I had recommended a simple numbering scheme for address slots such as net0/0 (unlike logical interfaces, such a numbering scheme would not be per-address-family -- that is, net0/0 would uniquely identify a slot on a phyint, not on an ill). It also is obviously a disjoint namespace, indicates the IP interface that it's associated with, and does not require explicit naming by the administrator. However, it's not clear how that scheme (or any scheme, really) expands to this address bundling mechanism that's needed for IPv6 that's being discussed. -- meem
