This commit removes the newer subprocess keywords that are not
available in 16.04/18.04. Tests work in 16.04 now too.

This PR adds a new io.netplan.Netplan dbus activatable service
that can be used to Apply a netplan configuration.

The main use-case for this is Ubuntu Core where we want to allow
snaps to control the netplan network configuration via an interface.

However we do not want to give access to systemctl to restart
systemd-networkd or network-manager. The dbus service will run
outside of the confinement will archive this goal.

The interface is probably useful beyond Ubuntu Core.

When using the NetworkManager renderer, the logic in
`NetplanApply.command_apply()` tells `nmcli` to disconnect the devices
previously populated by `netifaces.interfaces()`.

In addition, if an alias like "eth0:0" exists, it will show as a device
from `netifaces.interfaces()` but will not have a path in
`/sys/class/net`.

This means that the `devices` variable may have more devices than
actually exist due to the past `nmcli device disconnect` run or the
presence of one or more aliases.

The attached patch fixes this bug by tolerating all errors from `udevadm
test-builtin net_setup_link /sys/class/net/XXX` so that `netplan apply`
doesn't quit and leave the user without a network setup.

Fixes LP #1810043
Launchpad: https://bugs.launchpad.net/netplan/+bug/1810043



Signed-off-by: Nick Liu <deltik@gmx.com>

Signed-off-by: Mathieu Trudel-Lapierre <mathieu.trudel-lapierre@canonical.com>

Move validation and error reporting to their own files; also rework
the validation passes slightly to validate as much as we can of the
netplan grammar in parser passes (so we get context info for errors),
and leave the backend rules sanity checking only to after the parsing.

This way we'll get the error context, unless things are backend-specific
rules for which it's just hard to get the info (and not as helpful).

This reverts commit fe1fac85.

Partially revert the change for enabling systemd-networkd. network-online.target may not be depended on by anything on minimal images, but multi-user.target already is.

Signed-off-by: Mathieu Trudel-Lapierre <mathieu.trudel-lapierre@canonical.com>

Also rework how we setup for wifi tests in general to try and make them more
reliable.

Signed-off-by: Mathieu Trudel-Lapierre <mathieu.trudel-lapierre@canonical.com>

test_save, test_save_and_restore_with_dict: Ensure that we restore the
terminal state at the end of each test, as other tests rely on a clean
state to start. Previously, the tests would work the first time but fail
if re-run.

test_save: also compare self.terminal.orig_flags to the current
terminal flags, as opposed to a copy of itself (which would always pass)

Fixes LP: #1817660

This appeared in NM upstream as 648c256b9014198aac388097e410999c68c4b452; and
previous versions of NetworkManager do not seem to care, so land the changes in
general, without changes for previous Ubuntu releases.

Adjust tests so they still work with this change.

Signed-off-by: Mathieu Trudel-Lapierre <mathieu.trudel-lapierre@canonical.com>

Signed-off-by: Mathieu Trudel-Lapierre <mathieu.trudel-lapierre@canonical.com>

GLib apparently has some differences in the order in which members are
iterated in a GHashTable, which leads to things outputting from a GHashTable
to turn up in different order depending on the GLib version used to build
the generator. Skip these tests on codecov.io: we do have a pretty good test
coverage anyway.

Signed-off-by: Mathieu Trudel-Lapierre <mathieu.trudel-lapierre@canonical.com>

Signed-off-by: Mathieu Trudel-Lapierre <mathieu.trudel-lapierre@canonical.com>

Matching existing style in this file.

Init self.t so that we can simply test it in the `finally` clause.

`self.t` is only defined within the `try` block.  If an exception is thrown and the try bails out, `self.t` is undefined.  This causes the `finally` block to throw its own exception when it tries to call `self.t.reset`, and netplan crashes.

The original Changelog entry of Fri Dec 13 causes a build warning in rpmbuild. Based on the release date of 0.95 I updated the date in the changelog to be Fri Dec 14 which fixes the build warning

ex.
rpmbuild -ba netplan/rpm/netplan.spec 
warning: bogus date in %changelog: Fri Dec 13 2018 Mathieu Trudel-Lapierre <mathieu.trudel-lapierre@canonical.com> - 0.95

The source url is incorrect. the file name does not have the package name in it. its just version.tar.gz

The example block in "Properties for device type vlans:" on the "Reference"
site contains a key "address:" which doesn't exist. I suspect the key needed
is "addresses:".

LP: #1811802

We were not re-starting networkd/NetworkManager in the case we moved
from a configuration that created netplan files in /run to one that did
not need files there. Now we make sure we re-start the services also in
that case, by checking if there were configuration files before the
generation step. Fixes LP: #1811868.