I'm open to suggestions for test coverage. I did verify that this program:

#include <format>
#include <iostream>

int main() {
    volatile int v = 42;
    std::cout << std::format("{}\n", v);
}

was accepted before and rejected after the change, but it's hard to do "correctly fails to compile" tests internally.

Should your PR also close the issue #2427 ?

Also the issue says that

volatile double v = 42;
std::cout << std::format("{}\n", v);

works.

Should it be rejected too?

Looks good to me. After reading through the surrounding code, I saw that:

See also #2246

Should your PR also close the issue #2427 ?

Also the issue says that

volatile double v = 42;
std::cout << std::format("{}\n", v);

works.

Should it be rejected too?

I believe that behavior is consistent with the wording in the working draft; we should accept volatile float/double/long double/const char_type*/nullptr_t/cv-void* which are copied into non-volatile storage before formatting, otherwise only volatile program-defined class types whose formatter accepts volatile objects. I don't necessarily agree that the wording is sensible, but that is what it now says.

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

Thanks for handling this highly volatile issue! 🧪 🥼 😹