This fixes the case where repositories like debs-beta only have packages
in the main/common component (e.g., sid with only kernel packages).
Previously, the merge command would skip publishing if both utils and
desktop repos were empty, resulting in an incomplete repository.
Now we always publish at minimum the main/common component, ensuring all
distributions with any packages get properly published.
This commit implements a complete parallel repository management system
that allows building and publishing Debian repositories in parallel,
significantly reducing build time for multiple distributions.
- `update-main`: Builds common/main component once for all releases
- `update -R <release>`: Builds release-specific components in isolated DBs
- `merge`: Combines common + release-specific components into final repos
- Isolated databases (aptly-isolated-<release>) avoid locking during parallel builds
- Common component built once, not duplicated per release
- Release-specific components (utils, desktop) built independently
- Final merge combines all components with proper GPG signing
- Fixed GPG signing to target top-level Release files (dists/{release}/Release)
- Pool cleanup before publishing avoids "file already exists" errors
- Smart package import skips duplicates during merge
- Proper handling of empty repositories and missing components
- Improved error handling and logging throughout
1. update-main: Build common component (once)
2. update -R <release>: Parallel workers build release-specific components
3. merge: Combine all components and publish with GPG signatures
This enables GitHub Actions to run multiple release builders in parallel,
reducing total repository build time from hours to minutes.
Signed-off-by: Igor Pecovnik <igor@armbian.com>
- desktop images have half the world in their initrd (plymouth?)
- `bdinfo` shows reserved regions, which are hit depending on the size of initrd
- when reserved region is hit, u-boot says `** Reading file would overwrite reserved memory **`
- done by AI after looking at `bdinfo` reserved regions
- hard lesson: all `0x` hex have to be double quoted, otherwise `Wrong image format for "source" command`
- add note about Meco having conjured up `load_addr` (it's `loadaddr`); doesn't hurt
- while at it, remove copypasta/duplicate bootlogo/consoleargs stanza
- similar to `rkdevflash`, but for Mediatek devices
- also simpler; lk.bin & fip.img are produced by image build and directly used
- requires Rust+Cargo, so add those to hostdeps
- since this is a core extension, those will be included in all Docker images too
- which was bound to happen anyway since Rust in Linux Kernel is no longer an experiment
- extensions/mtkflash: sha1-based bin path, use fork & add `--no-erase-boot1`
- example invocation:
- `BOARD=radxa-nio-12l BRANCH=collabora RELEASE=trixie EXT=ufs,mtkflash MTKFLASH_TTYACM_DEVICE=1`
- u-boot stores env in UFS LUN 1
- use a 2mb offset (different from vendor/Collabora's) to avoid any env conflict
- point fw_printenv/setenv at it at the correct offset and size
- set SRC_CMDLINE, used in extlinux/EXT=u-boot-menu scenarios
- note about d4/d8/d16 irrelevance now as we have RAM autodetection
- note about where the blobs (libdram/libbase/lk.bin) live in GH: https://github.com/armbian/mtkbin
- this patches u-boot default addresses with the same values used in the boot-genio bootscript
- this enables booting extlinux with large kernels and initrd (eg: `EXT=u-boot-menu`)
- backport fixes from upstream u-boot for
- BTRFS (plus enable BTRFS and BZIP2 support for radxa-nio-12l)
- tested with `CARD_DEVICE=/dev/mmcblk1 ROOTFS_TYPE=btrfs BTRFS_COMPRESSION=zstd` (no UFS!)
- also with `EXT=ufs CARD_DEVICE=/dev/sdc ROOTFS_TYPE=btrfs BTRFS_COMPRESSION=zstd` (UFS!)
- 4k block size UFS for UMS: fixes block size issue (USB issues unhandled)
- Enable "bind" command for the Mediatek-standard USB Gadget Ethernet
- this way, one doesn't need to cripple UFS to be able to boot from SD
- USB didn't get the same treatment as it's slow and **not working**...
- ...and beyond me to fix; both OTG and Host mode are funky in u-boot...
- ...but seem to work fine in kernel
- IMPORTANT: keep in mind: "SDcard is NOT UFS (4k block)" and don't fall for traps!
- Network note:
- `setenv ethaddr xx:xx:xx:xx:xx:xx` & `saveenv` & `reset`
- required for network booting; mtk doesn't use random/cpu# MAC
- required for stable MAC in kernel (keep IP from changing, etc)
- Use TFA + libdram to obtain and set the actual memory size.
- Overwrite the memory size in the devicetree that is passed to the kernel with the actual size detected.
- All credit to Bartosz Bilas' work for the mt8188/mt8370.
- This doesn't really "use" TF-A/libdram though, those call bl33 with magic arguments which we read here.
- That is implemented by a patch in TF-A for the mt8195 platform.
- With this, bl2 will pass down the DRAM size it gets from libdram
down to bl33 (u-boot proper) via a magic location
- This is a rework of the implementation for another MTK plat (GRINN)
- genio: implemented complete bootloader build according to Collabora's scripts
- main difference from other similar stack is that TF-A builds final FIP, not u-boot
- thus I first disable ATF build initially, then build it in hook
- alternatively: we could build our own FIP, but that would diverge too much
- declare functions inside hook so hashing does all the versioning automatically
- write_uboot_platform() is a no-op; don't write bootloader to image
- ensure bootloader _builds_ across bookworm/trixie/jammy/noble/resolute
- keep in mind original code is for bookworm (tested to build and work on trixie)
- need to juggle the `-Wl,` prefix on LDFLAGS
- this is not mainline ATF and it does "stupid with flags"
- genio: u-boot: always clean optee & first atf build before building
- so one can switch build host with `DOCKER_ARMBIAN_BASE_IMAGE` (UFS...)
- genio: u-boot: using `armbian/mtkbin` repo
- outputs fip.img & lk.bin directly to output(/images)
- disk images do _not_ contain any bootloader
- small hack to get around stupid "From xx" parsing in Armbian patcher
- also: fix attribution in `0023-watchdog-mtk_wdt-Correct-maximum-timeout-from-15-to`
- Switched to the binman-atf-mainline scenario with the same rkbin DDR blob
rk3399_ddr_933MHz_v1.25.bin (TPL blob, as in the old working loader). DRAM init is correct, full
4 GB visible, Linux boots.
- U-Boot and DTS updated to v2026.01-rc5. Added MDIO/PHY node and phy-handle in rk3399-kobol-
helios64.dts, fixing “mdio node is missing”. Disabled i2s2 in the U-Boot DT to avoid rockchip-
i2s ... Could not register PCM.
- boot_targets in U-Boot config shortened (mmc1, scsi0, usb0, pxe, dhcp) to avoid env_buf ... too
small; however, the last built binary still shows the warning (needs rebuild with the shortened
list or enlarging the buffer).
- Boot flow: bootstd scans SD (mmc0) and eMMC (mmc1); SD boot works even without mmc0 in
boot_targets.
- Other changes: defconfig has CONFIG_ROCKCHIP_EXTERNAL_TPL=y to use the external DDR blob.
Boots.
The `DEST_LANG` variable affects rootfs cache content (locale is generated via
`locale-gen "${DEST_LANG}"` before cache is packaged), but it was not included in the
cache hash calculation.
This meant that changing `DEST_LANG` would not invalidate the existing cache,
potentially resulting in images with incorrect locale.
Include `DEST_LANG` in the `hash_hooks` calculation in `calculate_rootfs_cache_id()`.
Changes
- `lib/functions/rootfs/create-cache.sh`: Add `LANG=${DEST_LANG}` to the hash input
Result
Changing `DEST_LANG` now properly invalidates rootfs cache, ensuring the correct locale
is generated.
Build process generated numerous locale warnings like:
bash: warning: setlocale: LC_ALL: cannot change locale (en_US.UTF-8)
perl: warning: Setting locale failed
This happened because the Docker container and host environment were
configured to use `en_US.UTF-8`, but the rootfs cache may not contain
this locale (only `C.utf8` and whatever `DEST_LANG` specifies,
e.g. `en_GB.utf8`).
When `chroot_sdcard` runs commands inside the rootfs, environment variables
are inherited from the host/Docker, causing locale lookup failures.
Then use `C.UTF-8` locale for the build environment instead of `en_US.UTF-8`.
This locale is always available in rootfs immediately after mmdebstrap,
requiring no generation.
Changes
`lib/functions/host/docker.sh`: Remove `en_US.UTF-8` locale generation,
set `LANG=C.UTF-8` in container environment
`lib/functions/host/prepare-host.sh`: Change locale exports
from `en_US.UTF-8` to `C.UTF-8`
* tools/repository/extract-repo.sh: simplify extraction by copying directly from pool
Remove dependency on Packages index files. Instead of parsing package metadata
to find file locations, directly scan the pool/ directory structure and copy all
.deb files found in each component subdirectory.
This simplifies the code and makes it more robust since it doesn't rely on
index files being present or correctly formatted.
Signed-off-by: Igor Pecovnik <igor@armbian.com>
* Add helper script: recursively clean Armbian Debian package artifacts
* Update tools/repository/cleanup-debs.sh
Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
---------
Signed-off-by: Igor Pecovnik <igor@armbian.com>
Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
- turns out everybody was wrong, including me
- some (older?) ATF sources won't work, ever; thus
- introduce ATF_SKIP_LDFLAGS=yes to skip it completely
- introduce ATF_SKIP_LDFLAGS_WL=yes to only skip the `-Wl,` prefix
- this is for ATF's that pass flag directly to linker, not gcc
- artifact-uboot: hash atf-building code into artifact version
