cppcms

Build times (auto-updated)

• x86_64: 3m8s (2026-09-24 08:34 UTC, MAKEFLAGS=-j6)
• i486: 4m7s (2026-09-24 08:23 UTC, MAKEFLAGS=-j6)
• i686: 3m30s (2026-09-24 08:27 UTC, MAKEFLAGS=-j6)
• pentium4: 3m33s (2026-09-24 08:31 UTC, MAKEFLAGS=-j6)

---

• 2026-09-24: split into two packages, per user decision after an AUR comment thread (2025-08-21) where the user said they’d consider a beta package but not retarget this one to a beta. Upstream has since shipped a real 2.0.1 stable release (GitHub, 2024-11-02) – “C++17 compatible - stable release” – no longer a beta, so instead of a separate cppcms-beta package: this package (cppcms) now tracks the 2.x line (bumped straight to 2.0.1), and the old 1.2.1 build was forked off to a new package, memory/cppcms1.md, via a local git fetch <path> <branch> + git reset --hard FETCH_HEAD in the empty cppcms1 AUR repo – grafts full history without an actual push from this tree. cppcms1 conflicts/provides mirror this package so only one of the two can be installed at a time (upstream doesn’t version its own binary/header names between 1.x/2.x – /usr/include/cppcms, cppcms_tmpl_cc, cppcms_run are identical paths in both, so true side-by-side install isn’t possible without patching upstream’s install rules, not attempted).

• Source moved from SourceForge (last real release there: 1.2.1) to the GitHub tag tarball (artyom-beilis/cppcms, tag v2.0.1) – SourceForge only ever got as far as 2.0.0-beta2 (2020).

• 2.0.1 no longer needs python2: find_program(PYTHON NAMES python python2 python3) in CMakeLists.txt accepts plain python (Arch’s python3). Dropped the whole sed-based python2-forcing hack from build() that cppcms1/old cppcms needed.

• The booster/smart_ptr/sp_counted_impl.h two-arg allocate() call that forced the old -DCMAKE_CXX_STANDARD=17 workaround (see cppcms1’s history below) is gone from 2.0.1’s booster – genuinely fixed upstream, not just standard-pinned around. Kept the CMAKE_CXX_STANDARD=17 pin anyway though: upstream’s own CXXFLAGS still hardcode -std=c++11, but system ICU headers (auto-linked since icu-dev is present in this build environment; booster’s DISABLE_ICU_LOCALE cmake option auto-detects it) require C++17 – same conflict as before, just from the ICU side now rather than the allocator side.

• namcap (run via repo_build.sh’s own checkpkg step) flagged two real missing runtime depends that the original PKGBUILD never had, worth checking for on similar packages: icu (booster links it whenever icu-dev is present at build time, per above – not conditional at runtime) and python (the installed /usr/bin/cppcms_tmpl_cc script has a python shebang and is a genuine runtime dependency for anyone using it to build a cppcms project, not just a build-time tool). Both moved from makedepends-only to depends=.

• Also added license=('LGPL' 'MIT') correctly requires an installed license file for the “uncommon” MIT identifier (namcap: “Found 0/1 required license files”) – upstream ships COPYING.TXT/MIT.TXT at the source root but installs neither; added explicit install -Dm644 lines in package().

• 2026-09-24 08:49: user replied on the AUR comment thread themselves (comment 1086314), closing the loop on their own 2025-08-21 question above: “2.0.1 is out of beta. So I split the package into a cppcms1 which track 1.2.x and upgraded cppcms here to 2.0.1.” Self-answered, no reply needed from this tooling.

• 2026-09-24 build/release (with the icu/python/license fixes above): all four arches – i486, i686, pentium4, x86_64 – built, signed and published cleanly. Unlike python2 (x86_64-only in archlinuxaba, see cppcms1.md), plain python is available for all archs there, so no per-arch gap this time.

• AUR history: bb73fe3 (“using C++17 now”, the old 1.2.1-era HEAD) was not amendable – AUR’s server-side git hooks reject any non-fast-forward push (“denying non-fast-forward … hook declined”) even with --force-with-lease, so a local commit --amend + force-push attempt had to be recovered via git reset --soft bb73fe3 and redone as a plain new commit on top instead. History rewriting is simply not possible against AUR; always commit forward.

• Path: arch/maintained/cppcms

• Build failure during the 2026-07-07 “build all packages for x86_64” private-repo release pass (see repo_release_logs/maintained_cppcms.log).

• Missing-dependencies transaction listed cmake and python2, but only python2 actually fails resolution (error: target not found: python2); cmake is genuinely still in official extra (currently 4.3.4-1) and would resolve fine on its own. python2 itself was retired from the official Arch repos long ago and isn’t vendored in this tree.

• Same environment-drift class as memory/p2c.md / memory/bless.md / memory/bochs.md. Not fixed as part of this pass.

• 2026-07-30: unblocked – user built+published python2 to archlinuxaba directly (same ad-hoc-outside-this-tree pattern as exomizer/gconf/gtk-sharp-2/etc., see TOOLING_NOTES.md). i486/ i686/pentium4 still fail with target not found: python2 – expected, same chroot-scope limitation as dbmodel-qt4’s i686 leg (only the x86_64 build uses archlinuxaba-x86_64-build, which has the private repo layered in; plain extra-<arch>-build doesn’t see archlinuxaba at all).

  • x86_64 got past dependency resolution but hit a real compile failure, unrelated to python2: bundled booster (this project’s vendored Boost-alike, ancient/unmaintained upstream) calls the two-argument std::allocator::allocate(n, hint) overload in booster/smart_ptr/sp_counted_impl.h, deprecated since C++17 and removed in C++20 – and modern GCC (16 here) now defaults to a C++ standard newer than that. Same general class as TOOLING_NOTES.md’s “old C/C++ sources trip modern compiler/library changes” entries (modest, newsboat-og), but a removed stdlib API rather than a -Werror warning.
  • First attempt pinned -DCMAKE_CXX_STANDARD=14 to dodge the allocator removal – compiled further but then failed differently: system ICU’s headers (unicode/unistr.h, icu_78) require C++17 (u16string_view, std::is_same_v, auto template parameters, etc.), so C++14 was too old for the other side of the same translation unit. Fix that actually worked: -DCMAKE_CXX_STANDARD=17 (+ -DCMAKE_CXX_STANDARD_REQUIRED=ON) in build()’s cmake invocation – C++17 still has the deprecated-but-present two-arg allocate() and satisfies ICU’s C++17 requirements. pkgrel 3 ->
  4. Built and published cleanly on x86_64 after this.
  • Worth remembering as a general pattern: when an old bundled C++ library breaks under a newer default standard, check what its other dependencies (system libs it also compiles against, like ICU here) require before assuming the oldest standard that fixes the immediate error is the right one – there can be a narrow window (here: exactly C++17) rather than “anything old enough.”