/cc @mikioh @pmarks-net

and @mdempsky

I think this is working as intended. When you add a name to /etc/hosts, you're telling local resolvers to inhibit address lookups for the name. So if you define it as IPv4-only, then IPv4-only is what you get.

I understand your point, but I disagree. I think when an IPv4 host is added to a host file the system is being instructed to use that IP as a preferred IP. However if an IPv6 address is specifically requested and there isn't one present in the hosts file, but is via DNS, then it should be expected that the request will return the IPv6 address available via DNS.

This is how python acts:

$ grep example.com /etc/hosts
1.2.3.4 example.com
$

>>> socket.getaddrinfo("example.com", 80)
[(2, 1, 6, '', ('1.2.3.4', 80)), (2, 2, 17, '', ('1.2.3.4', 80)), (2, 3, 0, '', ('1.2.3.4', 80))]
>>> socket.getaddrinfo("example.com", 80, socket.AF_INET6)[0]
(10, 1, 6, '', ('2606:2800:220:1:248:1893:25c8:1946', 80, 0, 0))

It is also how ssh, ping6 etc work:

$ ssh -v example.com 2>&1 | grep Connecting
debug1: Connecting to example.com [1.2.3.4] port 22.

$ ssh -v -6 example.com 2>&1 | grep Connecting
debug1: Connecting to example.com [2606:2800:220:1:248:1893:25c8:1946] port 22.

$ ping6 example.com
PING example.com(2606:2800:220:1:248:1893:25c8:1946) 56 data bytes

So, according to getaddrinfo, the name is IPv4-only if you ask for both, but still has IPv6 if you ask for IPv6?

That's pretty strange, but I can't argue with emulating whatever getaddrinfo does.

I'm fine with relaxing the existing behavior of Lookup API by following some external configuration such as nsswitch.conf.

hosts: files dns // ask /etc/hosts then recursive servers
hosts: files // ask only /etc/hosts
hosts: dns // ask only recursive servers

Also fine to revisit the existing behavior. I can imagine it's pretty tough for people under some heterogenous environments to manage /etc/hosts, internal and public namespaces consistently.