Wrap the library in the ABI-versioning inline namespace

[Bronek]

Resolves #352.

Everything in fn and pfn now lives in inline namespace LIBFN_VERSION — spelled v0_0_9 today, staged to v0_1 at the 0.1.0 tag — so different library lines fail loud at link time instead of silently ODR-colliding. Three commits:

1. Stop the FWD and DEDUCED_RETURN macros leaking from fn headers. The standalone macro headers are folded into a guard-less bracketing pair fn/detail/macro_begin.hpp/macro_end.hpp (push_macro/define on entry, undef/pop_macro on exit — nesting-safe, and a user's own prior definition survives). Every per-header sentinel TU now #errors if FWD, DEDUCED_RETURN or ASSERT survives an include: the only macros a libfn header leaves behind are its include guard and LIBFN_VERSION. Tests that relied on the leak include the pair explicitly; the examples define their own one-line FWD, the idiom a user would write.

2. Add include/libfn_version.hpp, synced from VERSION. A new root header — below both fn and pfn, the only header either may reach outside its own tree — defining LIBFN_VERSION per branch (v0_0_9, or v0_0_9_cxx26 under the user-defined LIBFN_CXX26, reserved for the C++26 std::type_order mode). Never hand-edited: scripts/sync_versions.py derives both spellings from VERSION (now 0.0.9) exactly as it mirrors the packaging literals — v0_<y> once a 0.y line is tagged, v0_0_<z> on the 0.0 line, a SemVer prerelease appended (-dev → _dev), _cxx26 last. Packaging: exported CMake target libfn::version (FILE_SET, same EXPORT; carried by both libraries), Bazel :version; conan and nix pick the header up through their existing recursive copies.

3. Wrap fn and pfn in the inline namespace. All 35 namespace openings (the six fn/detail nested forms as namespace fn::inline LIBFN_VERSION::detail), each wrapped header self-containedly including <libfn_version.hpp>. A new pre-commit hook (scripts/check_namespace_wrap.py) refuses any unwrapped opening — a missed wrap compiles green while its entities silently escape the versioned ABI namespace, so the invariant is enforced textually. tests/fn/libfn_version.cpp pins fn::LIBFN_VERSION::… ≡ fn::… version-agnostically via the macro.

Empirical findings worth reviewing:

• Sortkey: GCC and MSVC spell the inline namespace in the pretty-printed type key; clang omits it. The version is deliberately kept in the key where the compiler spells it — truthful to the mangled reality and to what std::type_order will reflect — and the uniform insertion cannot flip any relative ordering, so copack canonical order is unchanged. Pinned per-compiler in tests/fn/detail/meta.cpp, version-agnostic via a stringized macro.
• Docs: doxygen dual-indexes inline-namespace members under both spellings (probed locally on 1.16.1), so the fn::… references in docs/ keep resolving; cmake/Docs.cmake additionally expands LIBFN_VERSION to the real spelling, parsed from the header. The docs CI job is the authority for the container's doxygen.

On the issue's open points: pfn receives the same treatment (one shared version header, one ABI story); docs are handled as above; the abigo baseline needs nothing here — the wrapped tree at the 0.1.0 tag becomes the baseline the 0.1 line's z-bumps classify against. The LIBFN_CXX26 wiring (CMake VALIDATE_CXX26, gcc-16 workflow lane, meta.hpp std::type_order) is deliberately deferred; its spelling is already reserved.

🤖 Generated with Claude Code

https://claude.ai/code/session_0198ikYYxtCqwNG54SS1b249

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[augmentcode]

🤖 Augment PR Summary

Summary: This PR introduces ABI versioning for libfn by wrapping all public APIs in fn and pfn inside inline namespace LIBFN_VERSION, so different library lines collide loudly at link time instead of via silent ODR mismatches.

Changes:

• Added include/libfn_version.hpp defining the LIBFN_VERSION macro (plus a LIBFN_CXX26 variant) and wired it into CMake and Bazel exports.
• Updated scripts/sync_versions.py to derive/update the namespace spelling from VERSION and keep packaging literals in sync.
• Wrapped all fn/pfn namespace openings with the versioned inline namespace; added a pre-commit hook (scripts/check_namespace_wrap.py) to enforce the invariant.
• Reworked internal macro handling: replaced standalone macro headers with fn/detail/macro_begin.hpp/macro_end.hpp bracketing to prevent FWD/DEDUCED_RETURN/ASSERT leakage.
• Adjusted documentation generation to stage a namespace-stripped copy of include/ for doxygen via cmake/StripNamespaceWrap.cmake and a new docs_stage_include target.
• Added/updated tests to validate the inline-namespace aliasing behavior and keep compiler-specific sortkey pins version-agnostic.

Technical Notes: Consumers should continue using fn::X/pfn::X; the inline namespace keeps source names stable while embedding the ABI version into symbol names.

_{🤖 Was this summary useful? React with 👍 or 👎}

[augmentcode]

Review completed. 2 suggestions posted.

Comment augment review to trigger a new review at any time.

[Bronek]

augment review

[augmentcode]

Review completed. 2 suggestions posted.

Comment augment review to trigger a new review at any time.

[Bronek]

augment review

[augmentcode]

Review completed. 2 suggestions posted.

Comment augment review to trigger a new review at any time.

[Bronek]

augment review

[sonarqubecloud]

Quality Gate passed

Issues
28 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[augmentcode]

Review completed. No suggestions at this time.

Comment augment review to trigger a new review at any time.