BindAddr may be "::" without "[]" (supported)

[jovandeginste]

(alternative is to not allow -bind ::)

Using :: and [::] is practically equivalent (https://www.ietf.org/rfc/rfc2732.txt) depending on some context.

[slackpad]

@sean- can you plz. look at this one wrt. to go-sockaddr?

[sean-]

:: and [::] are semantically equivalent in go-sockaddr and should be in Consul, too, but...

Consul doesn't need to use square brackets when establishing an IPv6 default address. We could/should make this change, but I'd suggest we align this with the other non-BC changes for 0.8. We could/should also interpret :: and [::] to their semantic values before performing any comparisons in command/agent/agent.go. Doing this correctly is a larger, more invasive change than I think we want to introduce in a point release because someone could be relying on :: as a way of binding to IN_ADDRANY for v6 addresses and bypassing the call to consul.GetPublicIPv6(). And for the record, it is possible to -bind=:: right now:

$ consul agent -dev -bind="::" -client='::1' -advertise='::1'

The issue is if you don't specify an advertise address, Consul dies because you can't advertise an endpoint that's the IPv6 equiv of -advertise=0.0.0.0/32.

I'm also pretty indifferent, but would guess given AWS's announcement re: IPv6 today, I think this is going to be a more common source of toe-stubbing in the near future. The question is, how far do we want to go in terms of fixing this? I think we do it right, but that involves some non-BC changes.

TL;DR: go-sockaddr will parse to the semantically equiv value, but Consul is currently using the string representations and there are corner cases where the behavior could be different.

[magiconair]

#2961 bans the use of an INADDR_ANY for the advertise address. Since we support "[::]" we should support "::" if they are equivalent. I can make this change once #3028 is merged. @sean- any objections?

[sean-]

@magiconair None, go for it.