oksh
Build times (auto-updated)
- armv7h: 4m28s (2026-09-16 12:01 UTC, remote, MAKEFLAGS=-j1)
- aarch64: 1m27s (2026-09-16 12:01 UTC, remote, MAKEFLAGS=-j1)
- armv6h: 8m50s (2026-09-16 12:01 UTC, remote, serial)
- i486: 1m13s (2026-09-16 05:47 UTC, MAKEFLAGS=-j6)
- i686: 1m14s (2026-09-16 05:48 UTC, MAKEFLAGS=-j6)
- pentium4: 1m13s (2026-09-16 05:50 UTC, MAKEFLAGS=-j6)
- x86_64: 1m5s (2026-09-16 05:51 UTC, MAKEFLAGS=-j6)
arch=('i486' 'i686' 'pentium4' 'x86_64' 'armv6h' 'armv7h' 'aarch64').
i486/i686/pentium4/x86_64 build in local archbuild chroots
(archlinuxaba-<arch>-build); armv6h/armv7h/aarch64 build remotely on
eurobuild4/5/14 (see config/remote_build.conf).
namcap flags Public Domain / BSD as non-SPDX license identifiers in this
PKGBUILD – upstream oksh’s actual licensing, not a packaging bug worth chasing.
2026-09-16: removed ancient generic arm subarch, rebuilt+published all 7 archs (7.9-2 -> 7.9-3)
User committed a packaging-only change dropping the never-buildable generic
arm entry from arch=() (nothing maps arm to a build command or a
config/remote_build.conf host, so it was always SKIPPED, never actually
building anything) – pkgrel bumped 2 -> 3 per usual convention, no functional
change to what actually ships. Rebuilt+published all 7 real archs via
repo_release.sh (no REPO_BUILD_ONLY_ARCH needed for the local 4; see below
for why the 3 remote ones needed a follow-up).
Remote ARM builds initially all 3 failed with a misleading “already building
something else” (empty holder label) on the very first attempt – looked
exactly like real board contention, but confirmed directly
(ssh <host> mountpoint -q /data) that all three boards simply had /data
unmounted (likely rebooted). This was a real ordering bug in
repo_build_remote.sh: the per-board build lock’s own mkdir under
REMOTE_BUILD_ROOT (/data/INSTALL/.building.lock) ran before the script’s
existing /data mount-ensure step, so on an unmounted board that mkdir fails
with plain ENOENT – indistinguishable from “lock dir already exists” by the
generic error path, and the cat .../holder fallback also fails (empty output),
producing the misleading empty-label message on 100% of retries regardless of
mounting. Fixed by moving the mount-ensure block before the lock acquisition
(see memory/TOOLING_NOTES.md’s 2026-09-16 entry). After the fix, one more
retry (REPO_BUILD_ONLY_ARCH="armv6h armv7h aarch64") built and published all
three cleanly – confirmed via repo_arch_status.sh showing all 7 archs
published (7.9-3).
2026-08-13: arch list expanded, republished
User expanded arch=() to add arm/armv6h/armv7h/aarch64 (pkgrel bumped 7.9-1
-> 7.9-2, packaging-only change per usual pkgrel-vs-pkgver convention).
Rebuilt+published for all 4 locally-buildable archs; the old 7.9-1 x86_64 build
got archived automatically by repo_publish.sh.