Disallow strings as input for wstring templates

[smat]

This change will prevent mixing character encodings by accident, by disallowing using strings as input for wstring templates.

The change will cause the following to be a compile error:

EXPECT_EQ(L"abc", format(L"{}", "abc"));

Use case: Using a wstring as a template with an UTF-8 string will mix UTF-32/UTF-16 and UTF-8 in the resulting string. When you try to convert the string to UTF-8, to_bytes will throw an exception.

fmt::WMemoryWriter out;
out.write(L"{}", "\u00F6");
const auto utf8_string = std::wstring_convert<std::codecvt_utf8<wchar_t>>().to_bytes(out.str());
std::cout << utf8_string << std::endl;

[vitaut]

Thanks for the PR!

This is somewhat related to #572. Instead of giving a compile-time error it might be better to convert a multibyte string into a wide one as if by calling mbrtowc. What do you think?

[vitaut]

Implemented a slightly different fix in ba01a5d. This gives a compiler error now:

fmt::format(L"{}", "\u00F6");

Thanks for catching the issue!

[hestad]

When will the next release of fmt be? We are waiting for this commit to find its way into spdlog :)