The only known incompatibility so far is the --iso ... option.
Therefore abort grml-debootstrap if DEBOOTSTRAP=mmdebstrap
and --iso ... option are used at the same time.

Closes: https://github.com/grml/grml-debootstrap/issues/120

Installation takes ages with lvm 2.03.02-1 if /run/udev isn't
available in the target system, causing messages like:

| WARNING: Device /dev/[...] not initialized in udev database even after waiting 10000000 microseconds.

Especially when grub package gets installed (which invokes
vgs and related tools) this takes forever and looks like a
failing installation.

This /run/udev bind-mount is a workaround required for Debian/buster's
lvm2, at least until #918590 is resolved.

No further changes needed

The "bullseye" release doesn't exist on the Debian repositories
as such yet, though it's supposed to appear at least once buster
(AKA Debian v10.0) is stable.

On 2019-02-12 there's the soft-freeze starting for the Buster release.
Accordingly default to buster as new default within grml-debootstrap.

Closes: grml/grml-debootstrap#117

In September 2018 the backports service was migrated from backports.org
to an official service within the Debian project and repositories (see
https://www.debian.org/News/2010/20100905).

This means http://backports.debian.org/debian-backports no longer exists
as such. Accordingly update the URL to point to the Debian repositories
(using ${RELEASE}-backports).

VirtualBox 5.2.22 is available in current stretch-backports, testing
(buster) and unstable. The URL for VBoxGuestAdditions_5.2.1-118918.iso
no longer works (results in 404).

grml-debootstrap used to happily create a target directory in /dev if
for instance one forgot to partition sda. To avoid such a mistake it
feels better to abort the installation.

Closes: grml/grml-debootstrap#99

Fix typo in comment, while at it

With the integration of cmdlinesopts.clp into the main script
this directory is no longer relevant.

kpartx (sometimes) does not cleanup the loop devices when deleting the
partition devmappings. This is a known bug, see Debian Bug #891077.

This is change is a possible workaround for that issue.

Now aborting grml-debootstrap is possible without leaving unneded loop
devices behind.

split finalize_vm into grub_install and umount_target

Quoting from the bug report:

| The timezone is currently setup in chroot-script:timezone(), but
| that function only sets up /etc/localtime but not /etc/timezone.
| The problem is that you end up with one set to Europe/Vienna and
| the other set to Etc/UTC.
|
| This can greatly confuse the system, due to some libraries/tools
| parsing the former, and other the latter.
|
| So that function should probably also update /etc/timezone, with
| the TIMEZONE value.

So write the $TIMEZONE setting to /etc/timezone.

Closes: #904607
Thanks: Guillem Jover <gjover@sipwise.com>

This is a big improvement, because we can now communicate with
the running VM, including executing processes inside the VM, like
running goss directly from inside the VM.

While at it also adjust travis/execute.sh to be able to run it
*locally* as well. We now don't have to necessarily push to Travis
CI, which also serves to be useful when developing new features.

Thanks: Christian Hofstaedtler <zeha@debian.org> for helping me with the serial console handling

This works fine on stretch.

We're starting a Docker container with Debian/stretch, installing
the Debian package build result (generated via travis.debian.net)
of grml-debootstrap inside it. Then execute grml-debootstrap,
installing to a VM file. Finally boot the generated VM file via
qemu.

We use stdio of qemu's serial console to check for the boot prompt
of the installed Debian system, using goss for running the actual
checks.

This fix was triggered because running --vmfile to create a virtual
machine image file caused the following error:

| device-mapper: reload ioctl on loop1p1p1 failed: Device or resource busy
| create/reload failed on loop1p1p1

This was caused because kpartx was run twice. The first run was used to test if
the Device-mapper was ready and the second run was actually add the partition
mappings. This used to fail under some circumstances.

While at it drop irrelevant stuff from .travis.yml.

Thanks Chris Lamb <lamby@debian.org> for https://travis.debian.net/ \o/

httpredir.debian.org forwards to deb.debian.org
nowadays, provided by Fastly and Amazon CloudFront.

Also drop tests/test_cmdlineopts.sh, it reports
success even though it can't even source the file…

Fix issues not yet found with shellcheck v0.4.4

This makes our life easier with shellcheck

Would be nice to avoid the shellcheck ignores overall, though I
don't want to break backwards compatibility with existing
configurations, so be as conservative as possible by ignoring
known incompatibilities (see esp. disable=SC2086).

Avoid the misleading error message if efi support isn't present,
by redirecting stderr to /dev/null as well.