Description of problem:
hostname -f returns <host> not <host>.<domain>

Version-Release number of selected component (if applicable):

systemd-libs-249.7-2.fc35.x86_64

How reproducible:

Always

Steps to Reproduce:
1. configure host line in nsswitch.conf to 
hosts: resolve [!UNAVAIL=return] files myhostname dns
2. hostnamectl --transient $(hostname -s) 
3. systemctl stop systemd-resolved
4. hostname -f

Actual results:

<host>

Expected results:

<host>.<domain>

Additional info:

Step 2 obviously does not include the domain name in the transient hostname. This is, apparently, what happens with dhcp network configuration and I would argue it is correct. Even though hostnamectl permits hostname to be set to a fqdn, linux has a limit of 64 characters, which is compatible with a DNS label limit of 63 characters, but not with a DNS fqdn limit of 253 characters.

nss-myhostname it seems, is trying to replace (dynamically) what would have traditionally be done via /etc/hosts, where the standard recommendation was to have the canonical name (fqdn) first with aliases after.

I think nss-myhostname should be appending the domain name when it looks up the interface addresses. This should be the first search domain in /etc/resolv.conf if it exists.

The hostname specification seems to be weakly standardized in the unix world, yet it needs to be consistent, at least on one installation. For example, readonly-root can use the hostname to specify a statefile held on an nfs server.  Currently, with the above configuration, hostname -f returns different results depending on whether systemd-resolved is running or not.

The above ordering in nsswitch.conf is consistent with the nss-myhostname(8) man page. The man page does contain a caveat regarding reverse resolution and a work around is to place 'myhostname' after 'dns' but this may not work if the real name server is not available (untested).