Reactivate Clang UBSan test coverage

[StephanTLavavej]

Fixes #3568.

This was briefly enabled for Clang 15 by #3452 on 2023-03-13, then disabled for Clang 16 by #3711 on 2023-05-18. After updating to Clang 20 by #5717 on 2025-09-15, we can re-enable this test coverage.

Note that tests/utils/stl/test/features.py disabled the configurations. We accumulated a couple of changes to tests/std/tests/VSO_0000000_vector_algorithms_floats/env.lst and tests/std/tests/usual_matrix.lst commenting out configurations, but I believe those changes were unnecessary. This PR updates features.py to enable the feature, uncomments the commented-out configurations, and doesn't change the large number of already-uncommented configurations across other files.

I audited our env.lst and related files, and I believe our UBSan coverage is reasonably comprehensive now. For the tests it's missing from, they either have good reasons (e.g. compile-only, /clr specific, incompatible with Clang for other reasons, etc.) or already have a blizzard of configurations where UBSan wouldn't be worth the extra cost.

Notably, UBSan has never been enabled for libcxx, and I am not attempting to do so here. libcxx configurations are extremely expensive (that's why we have only three). It may be reasonable to explore this in the future, but I don't want to spend developer and machine time on that now.

I'll leave Clang UBSan disabled for the MSVC-internal test harness (this was src/qa/VC/shared/testenv/bin/run.pl in MSVC-PR-457627, for future reference). I don't know if it works now, but it would provide negative value by burning a lot of machine time when GitHub is providing sufficient coverage of the STL's product and test code.

We've already merged a couple of fixes for bugs in vector<bool> and independent_bits_engine found by this impending PR:

• <vector>: The vector<bool> optimization for copy() performs a forbidden negative shift #5720
  • Fixed by @AlexGuteniev with Fix UB in vector<bool> by adding missing special case #5726
• <random>: independent_bits_engine performs forbidden full shifts #5719
  • Fixed by @MattStephanson with <random>: Prevent full-width shifts in independent_bits_engine #5740

⚙️ Commits

• Activate UBSan.
• Avoid duration_cast() UB, tracked by LWG-3921.
  • D:\GitHub\STL\out\x64\out\inc\__msvc_chrono.hpp:451:52: runtime error: signed integer overflow: 9223372036854775806 * 1000000 cannot be represented in type '_CR' (aka 'long long')
• Avoid UB caused by deliberate misalignment with #pragma pack.
  • D:\GitHub\STL\tests\std\tests\VSO_0000000_vector_algorithms_mismatch_and_lex_compare\test.cpp:177:9: runtime error: reference binding to misaligned address 0x002ec5cfe1bb for type 'short[1]', which requires 2 byte alignment
• Permanently avoid wmemcmp() UB, reported as VSO-2583476 "wmemcmp() triggers UB with a misaligned load".
  • C:\Program Files (x86)\Windows Kits\10\include\10.0.26100.0\ucrt\wchar.h:458:37: runtime error: load of misaligned address 0x7ff647c3b7bc for type 'unsigned long long', which requires 8 byte alignment
  • The offending line is: unsigned __int64 __v1 = *(unsigned __int64*)__s1;
• Properly detect UBSan in test code.
  • This is necessary because MSVC doesn't understand __has_feature. (I initially missed this because I was testing Clang only.)

[davidmrdavid]

LGTM, ty

[StephanTLavavej]

I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.