bacula7-client

Build times (auto-updated)

• x86_64: 2m3s (2026-10-14 05:00 UTC, MAKEFLAGS=-j6)

---

• 2026-10-14/15: ARM build (armv6h/armv7h/aarch64) not attempted – its sibling private/bacula7 (same src/findlib/fstype.c code path) hit a real, reproducible compile error on both armv6h and armv7h (narrowing conversion ... [-Wnarrowing], kills libbacfind.la) when attempted first; per the user’s explicit instruction to abandon both packages’ ARM builds on failure rather than chase it, bacula7-client’s own ARM attempt was skipped rather than run into the same wall. See memory/bacula7.md for the full failure detail. Still x86_64-only published; armv6h/armv7h/aarch64 were added to arch=() this same session but have never actually built successfully.

• Path: private/bacula7-client. Client-only Bacula 7.4.3 build (bacula-fd + bconsole, --enable-client-only --without-openssl). Alternative to the full private/bacula7-sd bundle’s bacula7-fd – provides=("bacula7-fd=${pkgver}"), conflicts=('bacula' 'bacula7-fd'). Ported 2026-10-13 from an old local build at /data/INSTALL/from_eurobuild3/bacula-client, renamed bacula-client -> bacula7-client (see memory/bacula7-sd.md for the shared naming rationale). pkgver stayed 7.4.3 even though a stray bacula-9.0.4.tar.gz and a stale .SRCINFO claiming pkgver=9.0.4 were also sitting in that old directory – the actual on-disk PKGBUILD was 7.4.3, so that’s what was ported; the 9.0.4 leftovers look like an abandoned upgrade attempt and were not touched or carried forward.

• Real bug: validpgpkeys was for the wrong key, left over from that same abandoned 9.0.4 attempt. The declared fingerprint (86260C02E82A8FC5CA5FB0638363575EFBD8D142, “Bacula 4096 Distribution Verification Key”) is real but signs the newer release line – the 7.4.3 tarball’s actual .sig is from a different, older key (2CA9F510CA5CCAF61AB529F59E98BF3210A792AD, “Bacula Distribution Verification Key”, no “4096”). Failed as gpg: ... FAILED (unknown public key 9E98BF3210A792AD) even after importing the declared key – the error names the key actually needed, not the one in validpgpkeys, which is the tell that it’s simply the wrong fingerprint rather than a keyring/chroot problem. Fixed by correcting validpgpkeys to the fingerprint above and re-running scripts/update_pgp_key_dir.sh. Confirms memory/TOOLING_NOTES.md’s existing note that a plain host-side gpg --recv-keys <fingerprint> (no chroot-keyring plumbing) is enough for local builds – once pointed at the right key it worked immediately.

• Build bug shared with bacula7-sd: src/console/conio.c is missing #include <termios.h>. Same fix, same patch file (02-conio-termios.patch, applies cleanly to 7.4.3’s conio.c too – identical surrounding lines) – see memory/bacula7-sd.md for the full explanation. Applied proactively here (in a prepare() this PKGBUILD didn’t have before) before ever attempting a build, since the bug is in the same upstream file bacula7-sd already hit.

• Dropped a stray duplicate sha256sums=() array that existed in the original PKGBUILD (one right after source=(), a second, different one after package() – bash just uses the last assignment, so only the second was ever actually in effect; the first was dead/confusing leftover from an updpkgsums retry). Kept only the one that matches the real checksum.

• Missing runtime depends (namcap E: findings, pkgrel 1 -> 2): the original depends=('openssl-1.0') was actually wrong for this package – --without-openssl means it’s never linked (namcap: “Dependency included, but may not be needed”). Real missing ones: libcap, acl, ncurses, zlib. Added python as optdepends (only used by the optional dvd-handler autochanger script, not worth a hard depends for a backup client).

• Built+published clean for x86_64 only (2026-10-13), REPO_BUILD_ONLY_ARCH=x86_64. i686/arm/armv6h not attempted this session – see memory/bacula7-sd.md’s i686 chroot-hang note before ever trying i686 for this one either.

• 2026-10-14: bumped pkgver 7.4.3 -> 7.4.7, pkgrel reset to 1. 7.4.7 is the final release of the 7.4.x branch (no 7.6/7.8 – upstream jumped straight from 7.4 to 9.0 in 2017); also bumped private/bacula7 (the renamed successor to bacula7-sd, see memory/bacula7.md) to the same 7.4.7 in the same session, so the two packages’ pkgvers are now aligned (previously 7.4.3 vs 7.4.5). New tarball’s .sig re-verified against the existing validpgpkeys fingerprint – same key, still good. The 02-conio-termios.patch still applies cleanly to 7.4.7’s conio.c. Built +published x86_64 only (REPO_BUILD_ONLY_ARCH=x86_64); old 7.4.3-2 entries auto-archived by repo_publish.sh.

2026-10-14: added armv7h/aarch64 to arch=() (armv6h was already there)

Same ARM-cluster-back-up request as [[bacula7]] this same day. arch=() here already had armv6h (and a stray, never-buildable 'arm' – left alone, not part of this request); added armv7h/aarch64. Built via REPO_BUILD_ONLY_ARCH="armv6h armv7h aarch64", i686 deliberately deferred for the same chroot-hang reason as bacula7.