Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

We've already got it, might as well use it.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

Use the more canonically accepted kernel file name. This frees up
kernel.img to be used for other things without causing unecessary
confusion, as kernel.img have either been the kernel file itself or the
UKI EFI binary.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

… arches

By popular demand.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

26.04 saw cdimage's structure change, so tarballs are in different locations depending
on the release.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

The files for checking what phase of the initrd
build had ocurred was only loosely utilized. This
change makes it more explicit.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

This should keep the initrd and kernel in their
appropriate lanes during builds.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

The behavior was backwards. Now it's correct.
We don't necessarily want cleaning to force failure
if umounting fails; the path could be unmounted for
a variety of reasons.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

dirmngr and gpg-agent love to hold on *forever*. Don't let them.
Cleaning actually cleans, now.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

pid in this case resolves to /proc/xxxx, which is invalid usage of kill.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

`echo` doesn't properly strip spaces in our variables like we need it to; since the
values are all already on new lines anyways we don't need to iterate in a loop anyways,
and we can just let IFS split our list and iterate over that directly.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

Because initrd_efi_image_{key,cert} have default values, the XOR check was always false.
Instead, the check must consider whether a value different from the default has been
set, and if one has been updated and not the other we should fail.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

Should improve error readability, which is important in this context as the user would
be the one making the mistake.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

The check in get_pull_commands is never actually hit, as commands.extend() will always
populate the pull commands with *something*. The failure will happen when the pull step
is actually executed.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

OLDPWD is previously set in a different function, and it's potentially possible that
build_tool could inherit this value instead of the correct one. To avoid the possible
(but very unlikely) edge case, make OLDPWD more explicit.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

This option supports building the kernel and its associated packages based on the
debian/rules makefile.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

It's possible for a user to specify:

initrd-addons:
  - foo/bar/
initrd-firmware:
  - firmware/baz/

In these cases, the path stripping will resolve these items to "", and the whole process
will silently fail. Trailing slashes should be stripped before the process begins to
handle this case. Multiple trailing slashes are unlikely to occur in this context, so
they are not handled.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

There are cases where custom configs should still be allowed, like with
kernel-ubuntu-debian-package. Kconfigs in that context would be
annotations fragments, and need to be handled in a specific way.

The docs have been clarified to be a bit more explicit with the
expectations associated with some options.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

This improves the readability of the primary function; now, most behavior
is sequestered into their own sections.

Traceability is still a little challenging, but at least the logic is
easier to follow now.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

Instead of passing comma-separated lists and parsing, we can pass quoted
lists to each argv and interpret in the "usual" sense. Drops many seds,
and use tr where appropriate.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

Should be easier to grok for the general reader.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

Extend the behavior of building debian package kernels to suppor the more
general use-cases commonly used by those building kernel debian packages.
The upshot is supporting DKMS packages.

Also resolves some small bugs or inconsistencies.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

Leverages the flexibility PartsErrors allows us and give suggestions to
the user for what to do instead.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

run() can be difficult to follow if you're not familiar with all the
moving parts. Include some explanation of how various options impact the
logical flow of each plugin.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

Reflects current state of changes.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

We always want to blow away any debs in this context.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

Rely on make install, should do the right thing now "usually".
Leave it on the user to know whether or not their tool's makefile respects make; make install.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

Includes broad changes to bring up-to-date with current plugin behavior and
expectations.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

Not single-quoting the value passed to kernel_build.sh results in quotes being stripped
from the values, but quotes are meaningful to the kernel's Kconfig system.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

It's quick enough that it doesn't need to be multithreaded and there's
the potential for race conditions, especially in more recent kernels
(6.18 has such an issue).

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

heredocs are fragile when embedded in yaml; to avoid annoyance, switch to some echos.

kernel test now checks the appropriate locations for built tools.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

and improve kernel tests.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

Avoid the nasty trickery trying to fetch cross-packages on Jammy and earlier.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

Switching to build-base: core24 means that we use platforms key instead of
architectures, and so the final snap package name is different.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

This test will probably never really succeed in CI due to requirements in the build
environment; we need binfmt_misc support in LXD to run the ARCH-specific binaries in
the chroot.

What really matters is that the premise works, and the native test achieves that goal.
If cross-building fails, it's probably for a reason unrelated to the plugin itself.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

This reverts commit 97f1f43.

cdimage has been updated so that resolute follows the same convention as other releases.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>

On amd64, the CPU microcode packages get removed if we rm -rf
${INITRD_ROOT}/lib/firmware *after* we configure the chroot, because
chroot_configure() installs the microcode packages! Clean firmware,
modules first. Then configure.

Signed-off-by: Dilyn Corner <dilyn.corner@canonical.com>