@DefaultRyan I tried using u8"literals" as you suggested but that doesn't seem to work in a way that will work with both C++17 and C++20. When compiled with C++20 those literals are char8_t string literals but char8_t is new to C++20 so I can't use it in C++17. I think I'll just leave it until we switch to C++20 and leave C++17 behind.

An approach I've taken elsewhere is to conditionally define the char type using the char8_t feature test macro:

#ifdef __cpp_char8_t
using char_type = char8_t;
#else
using char_type = char;
#endif

/azp run

/azp run

@kennykerr looks like you're doing changes from a different machine or the git config doesn't contain the mail associated with your account.

@walbourn Since all processes which share a library have separate memory they can store a flag to indicate does the current process uses UTF-8 or UTF-16. So if an app has UTF-8 strings, then there should be no conversion. Otherwise, conversion is required. Near the same way Nt and Zw functions work in Windows.

In case of WinRT where strings represented as handles and where the string lifetime is controlled by a reference counter, RAII could help: c_str should return a temporary struct which is implicitly convertible to char8_t/char, which does a conversion in the constructor and deallocation in the destructor if the flag reports that an app uses UTF-16. In case of UTF-8 no actions are needed.