config, test: make the BMI2 pext probe run, not just compile

[jll63]

Summary

Fixes the SIGILL failures in Drone build 492, stage 8 (Linux 24.04 GCC 13 32/64 UBSAN C++17): test_dispatch_minimal_cover_hash and test_hash_policies crashed with illegal operand/illegal opcode in the BMI2 pext-based dispatch path, on x86_64 only. The same failure showed up in build 488 on a different runner, and identically in build 492's stage 29 (macOS 10.15, Xcode 12.2/clang) - which rules out a GCC-13+UBSAN codegen bug and points at the actual cause: some CI runners pair a BMI2-capable compiler with a CPU that does not implement the instruction in hardware (pre-Haswell, or feature-masked under virtualization).

• config/has_bmi2.cpp was compiled with -mbmi2 as an obj target - a compile-only check. GCC/clang accept -mbmi2, and therefore emit pext, whenever the toolchain supports the extension; that says nothing about whether the machine running the build implements it. The probe answered "yes" on machines where the real tests then crashed.
• test/CMakeLists.txt had the identical latent bug one step further back: it gated -mbmi2 on CMAKE_SYSTEM_PROCESSOR alone, with no probe at all, so a developer building locally on pre-Haswell x86-64 hardware would hit the same SIGILL.

Also folded in a fix for an unrelated failure that showed up in the same CI run once the SIGILL was gone (Drone stages 9/10, Linux 24.04 GCC 13 32/64 UBSAN C++20/C++2b): a GCC 13 -Wstringop-overflow= false positive in test_hash_policies.cpp, x86_32 release only, confirmed present in the original failing build 492 too.

Changes

• config/has_bmi2.cpp: gave the probe a main() that calls pext64 on operands the optimizer can't fold away, so the instruction is genuinely executed.
• config/Jamfile: has_bmi2 target changed from obj to run, so check-target-builds executes the probe instead of only compiling it.
• test/CMakeLists.txt: replaced the CMAKE_SYSTEM_PROCESSOR guess with check_cxx_source_runs against the same has_bmi2.cpp source, so both build systems answer from one probe.
• test/test_hash_policies.cpp: ids_over()'s at vector now reserves before its push_back loop, like ids two lines below already does. GCC 13's -Wstringop-overflow analysis mis-traces growth of a default-constructed (capacity-0) vector through the deep inlining -O3 -m32 produces, and reports a memmove into a zero-sized region that cannot happen; starting from a reserved capacity removes the growth path its heuristic trips on.

The CMake change is strictly broader than the reported (b2-only) Drone failure. Flagging the tradeoff: check_cxx_source_runs unconditionally answers "no" under CMAKE_CROSSCOMPILING, where the old processor-only check would have answered "yes" blindly. Given the failure mode this guards against, "no" is the safe default.

Test plan

• b2 toolset=gcc libs/openmethod/config//has_bmi2 passes on this (BMI2-capable) machine
• b2 toolset=gcc libs/openmethod/test//test_dispatch_minimal_cover_hash libs/openmethod/test//test_hash_policies passes, and objdump -dC -M intel on the release-variant probe binary confirms a real pext instruction is emitted (not folded away)
• Simulated a probe runtime failure (forced non-zero exit): confirmed BMI2 pext : no is correctly detected and both real tests fall back to their no-op case instead of running pext
• b2 toolset=gcc address-model=32 ... still passes unaffected (BOOST_OPENMETHOD_HAS_PEXT is 0 there regardless)
• CMake: fresh configure shows Performing Test BOOST_OPENMETHOD_HAS_BMI2 - Success; built and ran both affected test binaries directly, both pass
• Reproduced the -Wstringop-overflow= failure directly with the exact CI compiler invocation, confirmed the reserve() fixes it, then re-verified across the full matrix it touches (cxxstd 17/20/2b, address-model 32/64, variant release/debug, undefined-sanitizer=norecover) - 0 failures
• b2 toolset=gcc libs/openmethod/test//quick still passes
• Confirmed against live Drone/GitHub Actions runs on this branch: stages 8, 29, 31 (the confirmed SIGILL cases) pass; stages 9, 10 were the only failures pre-fix and are addressed by the second commit

🤖 Generated with Claude Code

[cppalliance-bot]

An automated preview of the documentation is available at https://124.openmethod.prtest3.cppalliance.org/libs/openmethod/doc/html/index.html

If more commits are pushed to the pull request, the docs will rebuild at the same URL.

2026-09-21 17:21:58 UTC

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 92.30%. Comparing base (d44aa59) to head (159f10f).
⚠️ Report is 17 commits behind head on develop.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop     #124      +/-   ##
===========================================
- Coverage    93.51%   92.30%   -1.21%     
===========================================
  Files           22       27       +5     
  Lines         1695     2222     +527     
  Branches       504      679     +175     
===========================================
+ Hits          1585     2051     +466     
- Misses          66      112      +46     
- Partials        44       59      +15     

Files with missing lines	Coverage Δ
config/has_bmi2.cpp	100.00% <100.00%> (ø)

... and 1 file with indirect coverage changes

---

Continue to review full report in Codecov by Harness.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 1ba8152...159f10f. Read the comment docs.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.