- python3-pip implies a very old setuptools (which is system-wide and takes precedence)
- python3-dev implies python3-distutils (which is long deprecated)
- get-pip.py allows us to version pip in requirements.txt just like everything else
- in the end this fixes the conundrum with pylibfdt / dtschema on Jammy
- and, finally, the setuptools we specify in requirements.txt will be actually used
- refactor `prepare_host_binfmt_qemu()` out of `prepare_host_noninteractive()` and into `rootfs/qemu-static.sh`
- further split into more functions and return early to avoid deep nesting
- implement force import and load of qemu-arm for non-armhf capable arm64 hosts (incl magic numbers)
- enhance `deploy_qemu_binary_to_chroot()` & `undeploy_qemu_binary_from_chroot()`;
- add 2nd param "caller" for better logging/tracking
- does sanity-check and preserve existing binary if it exists
- explicitly deploy/undeploy for the 3 cases:
- image: moved undeploy from `post_debootstrap_tweaks()` into image build proper for consistency
- rootfs: was leaving trash behind (since post_debootstrap_tweaks never ran for rootfs), now properly undeploys
- initrd: was already fine, just added caller info
- added `arch-test` host dependency
- ensure `arch-test ${ARCH}` works during prepare-host
- > tl,dr: "can build 32-bit armv7 armhf using Apple silicon; can use rootfs cache cross-arch reliably"
Instead of having to add every new release to the list that does not support python2, inverse this if statement and only check if the build host is Debian bullseye or ubuntu jammy. Every release newer than those do not have python2. Older build hosts are unsupported.
Different build hosts have vastly different versions of setuptools and pyelftools depending on the host OS, e.g. Ubuntu 22.04 has setuptools v59 while the latest version at the time of this commit is setuptools v71.
Using Pip instead of APT to download these packages assures that all build hosts use the same version, removing some points of failures and inconsistencies.
doas is not compatible with sudo flags. The codebase was checked for
sudo-specific uses of this function, but none were found, all cases were
in the form of `sudo <command>`. Replacing it with `doas <command>`
yields the same result.
doas is not compatible with sudo flags. The codebase was checked for
sudo-specific uses of this function, but none were found, all cases were
in the form of `sudo <command>`. Replacing it with `doas <command>`
yields the same result.
This fixes the readability of colorized output of compile.sh when
running on a terminal with a light background. It uses the COLORFGBG
environment variable similarly to how the ip(8) command does.
Signed-off-by: Darsey Litzenberger <dlitz@dlitz.net>
- host_release (wanted_release_tag) was being re-computed from DOCKER_ARMBIAN_BASE_IMAGE
- DOCKER_ARMBIAN_BASE_IMAGE might have been changed from "<os>:<release>" if DOCKERFILE_USE_ARMBIAN_IMAGE_AS_BASE != no
- that would cause host_release to be "armbian-ubuntu-lunar-latest", not "lunar"
- which would cause python2/python2-dev to be installed, which would cause Dockerfile build to fail
- `USE_TMPFS=no` disables usage of generic tmpfs mechanism (still possibly used for rootfs/image building, which is unrelated), for last-resort cases
- use better/more descriptive `temp_dir_id`'s for kernel build than `k` (now `kernel_dest_install_dir`) and `kd` (now `kernel_debs_temp_dir`)
- specific image/dtb/headers packaging already had decent names, same for other .deb's
- replace `mktemp -d` with `mktemp -d --tmpdir "${temp_dir_id}-XXXXX"` in `prepare_temp_dir_in_workdir_and_schedule_cleanup()`, so we know what's using what in tmpfs
