however, i've found that if you want to swap around interface names that already exist with ifrename (e.g. if you want "eth0" to be "eth1", and "eth1" to be "eth0"), you need to rename one of them out to an alternate name first, because it can't rename interfaces to an existing name. I get around that particular problem by giving my interfaces nice clean names that aren't auto-assigned names anyway (e.g. "wired" and "wireless", or "lan" and "wan"). This kind of semantic naming is actually pretty useful for scripts which do things to the interfaces as well.

As for doing everything with udev (and no extra packages), i see your point. It feels like a cleaner solution. However, i'm scared of tampering with udev config files because the udev maintainer seems to change them so often. I've only got one simple change to my udev config on a machine running etch (changing the mode of /dev/nvram from 0660 to 0640), and i find myself constantly having to juggle config file changes when udev upgrades come through.

But I haven't tried creating a separate config file that lives outside the shipped config files, like you've done here. i like that because it's a more modular and maintainable arrangment. But reading man udev, i can't figure out which rule takes precedence in the event that two rules match. Is it the first matching rule discovered lexically in /etc/udev/rules.d/ or the last matching rule?