xxdiff-git
Build times (auto-updated)
- x86_64: 17m9s (2026-10-10 09:07 UTC, MAKEFLAGS=-j6)
Build times (auto-updated)
- x86_64: 2m27s (2026-10-05 23:40 UTC, MAKEFLAGS=-j6)
- Tracked under
arch/maintained/xxdiff-git. PKGBUILD is ours (Contributor: Andreas Baumann), not a vendored AUR one –build()runsmake -f Makefile.bootstrap(callsqmake6 -o Makefile.qmake xxdiff.pro) thenmake.
2026-09-22: armv6h build fails – bad atomic linker arg, not in our source
Final link line has BOTH the correct -latomic AND a bare trailing atomic
token:
... -lpthread -latomic -lGLX -lOpenGL atomic
g++: error: atomic: No such file or directory
g++ treats the bare atomic as a filename to link, which doesn’t exist. Checked
upstream xxdiff.pro/Makefile.bootstrap directly (fresh clone) – no mention
of “atomic” anywhere in the source tree at all. This means the bad token is
being generated by qmake6 itself on that armv6h board, almost certainly a
Qt6 mkspec quirk for the armv6l target (some mkspecs append a bare atomic to
QMAKE_LIBS_PRIVATE-style variables instead of using the -l-prefixed form
qmake normally auto-adds).
Not fixed this pass – didn’t want to guess-patch the generated Makefile without
being able to verify on real ARM hardware (each attempt costs a real build cycle
on a slow board). If revisiting: try sed-fixing the qmake-generated Makefile
in build() after the bootstrap step (strip the bare atomic token, or
override QMAKE_LIBS_PRIVATE), or check whether a newer/older qmake6 on that
board avoids the bug.
2026-10-10: x86_64 rebuild+republish, other archs failed on infra not the package
r552.b455827-1 built+published cleanly on x86_64. i686/pentium4 failed with
mirror.archlinux32.org connection timeouts (local mirror sync issue);
armv7h/armv6h/aarch64 failed because eurobuild3 (shared NBD /data server for
those remote boards) was unreachable – known infra dependency, see
TOOLING_NOTES.md. Not the armv6h atomic bug above (never got that far this
pass) – just couldn’t reach the boards at all.