git-wd40

Build times (auto-updated)

• armv7h: 4h59m (2026-09-13 16:02 UTC, remote, MAKEFLAGS=-j1)
• aarch64: 1h32m (2026-09-13 16:02 UTC, remote, MAKEFLAGS=-j1)
• i486: 7m32s (2026-09-13 06:26 UTC, MAKEFLAGS=-j6)
• x86_64: 10m58s (2026-09-13 06:49 UTC, MAKEFLAGS=-j6)
• armv6h: 9h13m (2026-09-13 16:02 UTC, remote, serial)

---

• Path: arch/maintained/git-wd40

• 2026-07-06: built/published 2.55.0-1 to the private archlinuxaba repo for i486 and x86_64 only. i686 and pentium4 failed identically while linking the contrib/credential/libsecret helper:

  /usr/bin/ld: .../libsecret-1.so: undefined reference to `g_variant_builder_init_static'
  

  i486 and x86_64 built the same step fine, so it’s not a PKGBUILD bug.

  • 2026-07-06 diagnosis was wrong: blamed a stale local chroot (extra-{i686,pentium4} drifting out of sync) and prescribed a pacman -Syu fix. Retried 2026-09-12 after actually resyncing both chroots (arch-nspawn .../root pacman -Syu – i686 reported already up to date, pentium4 genuinely updated several packages) – identical failure, unchanged. Real root cause, confirmed directly: both chroots have glib2 2.80.0-2.0 installed alongside libsecret 0.21.7-1.1, and that libsecret build references g_variant_builder_init_static, a GLib symbol not present in that old glib2. This is an upstream Arch32 core/extra snapshot that’s internally inconsistent for these two archs specifically – same systemic i686/pentium4-lags-behind pattern as notion3’s texlive-bin/texlive-basic skew and gconf’s glib2-devel split gap (all three hit in the same 2026-09-11/12 session). Not fixable by resyncing our own chroot – the inconsistency is upstream’s, not ours. Nothing to do here; revisit once Arch32’s extra/core for i686/pentium4 catches up (check glib2’s version there, same as the other two cases).
  • i486/x86_64 are already published at 2.55.0-1; i686/pentium4 stay unpublished until upstream catches up.

• Its source=() includes a git+https://... VCS entry, which extra-<arch>-build clones into a local git/ directory in the package dir (~300MB) — this isn’t covered by repo_build.sh’s generic post-build cleanup (VCS checkout directory naming varies too much per package to safely generalize), so remove it manually after a build: rm -rf git/ (confirm untracked first: git ls-files --error-unmatch git).

2026-09-12/13: armv6h doc-skip + armv7h PATH fix published, recovered from a mid-session data-loss incident

Both fixes from this session confirmed working in a real full release:

• armv6h’s arch-conditional doc/man skip (avoids the xmlto/ xsltproc OOM on that 423MB board) – published clean at pkgrel=2.
• armv7h’s pod2man: command not found – fixed generally in remote_build.sh itself (adds core_perl/vendor_perl/site_perl to PATH, since non-interactive ssh -n skips the login-shell profile script that normally does this) – published clean.

Mid-session incident, since fixed structurally: a second concurrent repo_build_staged.sh invocation for this same package wiped the first invocation’s already-built-but-not-yet-copied-back i486/x86_64/aarch64 output (the copy-back-to-pkgdir step only runs once, at the very end) – lost a full rebuild cycle for those three archs before the mistake was caught. Root cause: launched a second repo_release.sh maintained/git-wd40 while the first was still running, without checking. Fixed structurally, not just “be more careful”: repo_build_staged.sh now has its own per-package mkdir-atomic lock (mirrors the remote per-board lock added the same session) – see memory/TOOLING_NOTES.md’s 2026-09-13 entry. Final clean rebuild after the fix published i486/x86_64/armv6h/ armv7h/aarch64 all at pkgrel=2; i686/pentium4 still blocked by the unrelated upstream glib2 lag documented above.