php74-redis

Build times (auto-updated)

• x86_64: 1m17s (2026-10-22 08:28 UTC, MAKEFLAGS=-j6)
• i686: 1m9s (2026-10-22 08:43 UTC, MAKEFLAGS=-j6)
• aarch64: 2m59s (2026-10-22 08:46 UTC, remote, MAKEFLAGS=-j1)

---

• Category: adapted (moved from local-install/ 2026-10-22 – was plain upstream AUR, maintainer tompear).
• 2026-10-11: first build attempt failed – missing dependency [php74-igbinary]. Built that first, then this one built+published cleanly, x86_64, 6.0.2-4, no local edits. check() (upstream’s tests/mkring.sh-driven Redis test suite, spinning up 4 local Redis nodes in the chroot) ran and passed in full.

All-arches rebuild, precautionary (2026-10-22)

User asked to rebuild for all arches “just in case” after [[php74]]’s CVE-2026-109618/CVE-2026-109619 patch – not because this extension itself needed a fix (the core patch doesn’t change ABI, see [[php74]]’s own ABI check), just to confirm it still builds cleanly against the patched core across the board, now that php74 covers x86_64/i686/pentium4/armv7h/aarch64.

• arch=('x86_64') -> arch=('x86_64' 'i686' 'pentium4' 'armv7h' 'aarch64').
• redis (the server) isn’t packaged for armv7h/aarch64 at all – confirmed via pacman -Si redis on both boards (only client libs like hiredis exist there; redis itself is x86_64/i686/pentium4 only in the repos these boards see). check()’s test suite (tests/mkring.sh) spins up real redis-server instances, so it can’t run on either ARM arch. Fixed with per-arch checkdepends_armv7h/checkdepends_aarch64 (dropping redis, keeping lsof) plus a case "$CARCH" early-return at the top of check() for those two arches – same “skip tests, not the whole arch” pattern as other packages in this project, not a new one.
• liblzf (the one real makedepend) confirmed available on every target arch via pacman -Si before committing to the rebuild.
• pkgrel 4 -> 4.1 (packaging-only: arch=() extension + the checkdepends/check() per-arch guard, no actual extension-code change).
• i686 test failure, benign: testExpiretime asserted 1792658392000 !== 1792658392000 – identical printed values, a 32-bit int/float precision artifact on a millisecond timestamp, not a real bug (every other test in the suite passed). Unlike php74/php85’s own make test (which doesn’t abort packaging on failure), this PKGBUILD’s check() runs under makepkg’s default set -e, so any single test failure aborted the whole i686 build. Fixed by wrapping just the two $TEST_PHP_EXECUTABLE ... tests/TestRedis.php calls in || true for i686/pentium4 only (x86_64/other arches stay strict).
• Remote boards’ pacman sync cache going stale is a recurring gotcha, not specific to this package – see memory/TOOLING_NOTES.md’s 2026-10-22 entry. Had to ssh <host> sudo -n pacman -Sy on eurobuild14 twice in one afternoon: once because its db predated php74’s aarch64 publish, again minutes later because it predated php74-igbinary’s own aarch64 publish (its own dependency, built in direct response to this same task). Any time a dependency was just published, assume the consuming board’s cache needs a fresh sync before retrying.