Two "su pmos -c 'git config --global user....'" commands were added to
the before_script in .gitlab-ci.yml. They work fine with the regular
gitlab runners, but fail with the custom gitlab runner that runs the
qemu test. Simply allow the command to fail, because it isn't needed on
the custom gitlab runner, as it doesn't run related tests.

Fixes: 16e2d3c7 ("gitlab-ci.yml: set git user/email (!1848)")

Right now we explicitly add ttf-droid to /etc/apk/world, and force
installation of it on every postmarketOS installation. But there really
is no reason to force this from pmbootstrap:

  - If ttf-droid is really required for every postmarketOS installation,
    it should be a depends of postmarketos-base.

  - Even with this change it is still installed on every device,
    because postmarketos-splash depends on it.

  - There is no need for fonts when using the "none" UI.

For now, the only difference of this change is that we no longer add
ttf-droid explicitly to the APK world. Instead, it is installed as
dependency of postmarketos-splash.

At the moment we assume that everyone running QEMU is either using
ALSA, or has the ALSA PulseAudio plugin configured on the host system.
(Since we run QEMU outside of the chroot at the moment, configuration
files are read from the host system instead of the chroot...)

Other distributions have much wider support for PulseAudio,
so not everyone will actually have the ALSA PulseAudio plugin configured.
In that case, it is better to use QEMU's PulseAudio backend since that
does not require any configuration. It also lets us drop the alsa-lib
fork, since that was only needed for loading the ALSA PulseAudio plugin.

In general, it seems difficult to detect which audio backend the user
wants to run (if any). Selecting the wrong one results in ugly warnings
when running QEMU. So let's not assume any by default, and add a
--audio option instead which accepts one of QEMU's audio backends
(alsa, pa or sdl).

Add a shortcut for "git pull --ff-only" in all repositories cloned by
pmbootstrap (currently pmaports and aports_upstream, new pmdevices
repository coming soon).

'pmbootstrap pull' will only update the repositories, if:
* they are on an officially supported branch (e.g. master)
* the history is not conflicting (fast-forward is possible)
* the git workdirs are clean

Otherwise it shows the user a descriptive message about what to do. The
list of supported branches is only "master" right now, and will be
extended in later commits, so we can have a stable branch for pmaports
based on Alpine's releases. More about that in the project direction
2020 issue.

Closes: #1858

Prevent upcoming tests for "pmbootstrap pull" from failing, when trying
to create a git repository.

Get the path to the repository, which is either the default one in the
work dir, or a user-specified one in args.

Remove a legacy code path from back in the day, where pmaports were in
the same repository as pmbootstrap ([1]). Nowadays, pmaports are always in a
git repository, so this is not necessary anymore. If git fails at this
point, crash hard and don't carry on with an empty string.

This is in preparation for using rev_parse in new code paths, so we
don't need to check the return value of rev_parse.

[1] 39548835 "Write custom os-release (closes #324) (#439)"
    https://github.com/postmarketOS/pmbootstrap/pull/439

When defining a new kernel subpackage with a "-" in it
(e.g. $pkgname-kernel-mainline-modem), then pmbootstrap is unable to
find the function that builds the subpackage:

ERROR: Could not find subpackage function, no line starts with 'kernel_mainline-modem() {'

This is because it assumes that a $pkgname-kernel-<name> subpackage
is built by a kernel_<name> function, but this does not have to be the case.

We should really respect the name of the subpackage function that
is specified when defining the subpackage, but unfortunately it is
stripped away in cut_off_function_names().

For now let's fix this by replacing "-" with "_", but ideally the
APKBUILD parser API should be refactored to expose the subpackage
function in the future.

Add a test for APKBUILD variable parsing as well

Fixes: 9f1dd64f ("install/partition: allow specifying a different boot partition start (!1856)")

This adds a new deviceinfo parameter, 'boot_part_start' which accepts an
int and indicates the number of sectors from the start of the drive to
place the boot partition.

The librem5 devkit (and actual phone) u-boot has grown beyond the 2048
sector space previously before the boot partition, so this is necessary in
order to boot pmos on this device.

Install rust (rustc) in native chroot, if it is in the APKBUILD's
build dependencies. Add a test to verify that crossdirect + rust works
as expected.

Closes: #1861

Rust packaging is new and still a bit weird in Alpine and postmarketOS.
As of writing, we only have one package (squeekboard), and use cargo to
download the source of all dependencies at build time (several git
repositories!) and compile it. Usually, this is a no-go, but at least
until this is resolved properly, let's cache the downloads as suggested
in: https://doc.rust-lang.org/cargo/guide/cargo-home.html

Related: #1861

When initializing the building chroots, some symlinks are created to
/mnt/pmbootstrap* dirs. With upcoming rust-related symlinks, they will
point to subdirs in /mnt/pmbootstrap-rust, such as
/mnt/pmbootstrap-rust/registry/cache. Create this directory structure if
it does not exist.

Do not install git in the native chroot and use it from there. Remove the
chown_to_user argument from pmb.helpers.git.clone(), the resulting dir
is now always owned by the user. While at it, refactor the function and
display the clone URL.

Previously we had cloned aports_upstream (from Alpine) with
chown_to_user=False (legacy) and pmaports with chown_to_user=True.
pmb.helpers.git.rev_parse() would only work after chown_to_user=True.

Check if git is installed in "pmbootstrap init", and remove the same
check from rev_parse(). Add a new work dir version, that checks for git
and changes ownership of already checked out aports_upstream to the
host system's user.

When creating a new work dir, create cache_git instead of cache_http.
cache_http is created on demand already, with proper permissions. But
cache_git must be created, otherwise pmb.helpers.mount.bind will create it
as root.

This is in preparation for the "pmbootstrap pull" feature, as it allows
using the host system's git in all new code paths. We will be able to
handle repositories even if they were cloned outside of the work dir
(which we do in a few CI scripts for example).

Related: #1858

Remove /home/pmos/git from pmb.config.chroot_home_symlinks, so
pmb.chroot.init.init() will not change the ownership of the
$WORK/cache_git dir.

The symlink was added two years ago in [1] as compatibility to previous
behavior, because the mountpoint of cache_git was changed from
/home/user/git to /mnt/pmbootstrap-git. Forcing the ownership of that
dir to be the pmos user inside the chroot made sense, because the only
cloned repository (aports_upstream) would also be owned by the pmos
user. With the upcoming change to have all cloned repos owned by the
host-system's user, this is just confusing.

The /home/pmos/git symlink does not seem to be used by anybody, it is
not mentioned in the pmbootstrap code and in the wiki.

[1] b4dd7a89 "Close #709: Improve user creation (#725)"
    https://github.com/postmarketOS/pmbootstrap/pull/725

Use Alpine's primary mirror instead of the mirrored version at github
for new clones. A follow up commit will display the clone URL.

This should have already happened, but probably broke when splitting
pmaports into a separate git repository.

Prepare to require "git" in a follow-up patch.

The APKBUILD reference for "provides" [1] is not worded clearly; but
after reading it over and over again, my understanding is:

* package with provides='foo=1.2' will be automatically installed if
  user requests installing "foo"
* package with provides='foo' (without version) will NOT get
  automatically installed if user requests installing "foo"

For pmbootstrap, this means, that we must not attempt to build a package
where the pkgname mentioned in provides matches what we are currently
resolving, unless there is an equals sign in the provides entry.

Fixes: #1862, pmaports#404

[1] https://wiki.alpinelinux.org/wiki/APKBUILD_Reference#provides says:

"List of package names (and optionally version info) this package
provides.

If package with a version is provided (provides='foo=1.2') apk will
consider it as an alternate name and it will automatically consider the
package for installation by the alternate name, and conflict with other
packages having the same name, or provides.

If version is not provided (provides='foo'), apk will consider it as
virtual package name. Several package with same non-versioned provides
can be installed simultaneously. However, none of them will be installed
by default when requested by the virtual name - instead, error message
is given and user is asked to choose which package providing the virtual
name should be installed."

Alpine's APKBUILD for gcc supports a new language that we don't need, so
disable it just like the irrelevant ones.

Related: pmaports!839

Changes:
  * Added the ALSA_PLUGIN_DIRS environment variable otherwise it tries
    to load the plugins of the host pc and they fails.
  * Added the /usr/lib/pulseaudio path to the library-path argument
  * Added the audiodev and soundhw parameters to the qemu command
  * Added the required packages necessary to be installed in the chroot

When git is installed, the Linux kernel build system appends suffixes
to the Linux kernel version based on the local repository state.
This means that we get different kernel versions when building the
package normally or with envkernel.sh.

This causes problems with Alpine's "installkernel" script,
which detects the kernel flavor based on the kernel version.
e.g. when building linux-postmarketos-qcom-msm8916 with git installed
the kernel image is installed to /boot/vmlinuz-g556e2c21dbac-dirty
where pmbootstrap (and anything else) cannot find it.

Two changes are necessary to prevent the kernel build system from
appending Git version information:

  1. Set LOCALVERSION= - this is done by this commit.
  2. Unset CONFIG_LOCALVERSION_AUTO
     - this needs to be done in each kernel config (when needed)

Adjust to new text that "file" outputs for ARM linux kernels:
	"ARM OpenFirmware FORTH Dictionary..."

Make test_bootimg_kernel pass again, which was failing with:
	AssertionError: assert 'heimdall-isorec' in 'File is not an Android ...

The kernel build system tries to use find -printf when
CONFIG_INITRAMFS_SOURCE is chosen, but busybox find doesn't support
that.

Fixes: #1849