Memory corruption on Windows with parametrized tests

[SuperFola]

I'm investigating a problem that popped up on my CI today ; everything was fine on GitHub CIs on December 9th, not anymore, but I can replicate the issue on an ARM Windows 11 (up to date as of 2025/12/11, today) with VS 2022 and MSVC.

Basically running the following test:

void tt(const std::string& folder, std::function<void(const std::string&)>&& run)
{
    boost::ut::test(folder) = [&] {
        for (const std::string& name : {
            "11111111111111111111111111111111111111111111111111111111",
            "2222222222222222222222222222222222222222222222222222222222",
            "2222222222222222222222222222222222222222222222222222222222",
            "333333333333333333333333333333333333333333333333333333333333"})
        {
            run(name);
        }
    };
}

boost::ut::suite<"Main"> main_suite = [] {
    using namespace boost::ut;

    tt("main", [](const std::string& name) {
        should("parametrized " + name) = [] {
            expect(true);
            expect(false);
        };
    });
};

Will output

Running test "ain"...
  Running test "¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦"... FAILED
in: Z:\ArkScript\Ark\tests\unittests\Main.cpp:30 - test condition:  [false]


  Running test "¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦"... FAILED
in: Z:\ArkScript\Ark\tests\unittests\Main.cpp:30 - test condition:  [false]


  Running test "¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦"... FAILED
in: Z:\ArkScript\Ark\tests\unittests\Main.cpp:30 - test condition:  [false]WARNING test '¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦' for test suite 'Main' already present

  Running test "¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦"... FAILED
in: Z:\ArkScript\Ark\tests\unittests\Main.cpp:30 - test condition:  [false]


  Running test "¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦"... FAILED
in: Z:\ArkScript\Ark\tests\unittests\Main.cpp:30 - test condition:  [false]

on Windows, while it would output

Running test "main"... 
  Running test "parametrized 11111111111111111111111111111111111111111111111111111111"... FAILED
in: /ArkScript/Ark/tests/unittests/Main.cpp:30 - test condition:  [false]


  Running test "parametrized 2222222222222222222222222222222222222222222222222222222222"... FAILED
in: /ArkScript/Ark/tests/unittests/Main.cpp:30 - test condition:  [false]


  Running test "parametrized 2222222222222222222222222222222222222222222222222222222222"... FAILED
in: /ArkScript/Ark/tests/unittests/Main.cpp:30 - test condition:  [false]WARNING test 'parametrized 2222222222222222222222222222222222222222222222222222222222' for test suite 'Main' already present

  Running test "parametrized 2222222222222222222222222222222222222222222222222222222222"... FAILED
in: /ArkScript/Ark/tests/unittests/Main.cpp:30 - test condition:  [false]


  Running test "parametrized 333333333333333333333333333333333333333333333333333333333333"... FAILED
in: /ArkScript/Ark/tests/unittests/Main.cpp:30 - test condition:  [false]

on Linux/MacOS.

I don't see any problems with the above code regarding lifetimes, nor does the compiler (running with -Wall -Wextra -pedantic -Wstrict-aliasing -Wshadow -Wconversion on Linux, alongside ASAN and UBSAN, and /W4 /MP4 /EHa on Windows).

Am I doing something wrong with ut or is it Windows and MSVC being derps?

[SuperFola]

I managed to somewhat replicate the bug on https://godbolt.org/z/4cWT6aPqG (ut 1.1.9) and on https://godbolt.org/z/j767j71Ye (ut 2.3.1), based on what was said in #637. Clang doesn't exhibit the segfault nor the memory corruption, but MSVC has the memory corruption happening even on godbolt.org.