constexpr checks of format strings

[vitaut]

Add a check_format constexpr function for performing a compile-time format string check and a section explaining why such check cannot be incorporated in the format function itself at the moment. As discussed in #492 and the main problem is the lack of ability to handle the format string in constexpr context when other arguments are not constexpr.

Also Zach Laine suggested to mention the benefits of "compile-time strings":

But the other part of what I want is for people to see how nice the interface would be if you could treat the format string as a constant expression.

[vitaut]

Potentially relevant discussion about constexpr function parameters: https://groups.google.com/a/isocpp.org/d/msg/std-proposals/RdAK-0RyiY0/F3h-qF_jWbIJ

[vitaut]

It is possible to access the format string and argument types at compile time using UDLs:

#include <string>

template <char... Chars>
struct formatter {
  template <typename... Args>
  constexpr std::string operator()(Args...) const;
};

template <typename T, T... Chars>
constexpr auto operator""_format() {
  return formatter<Chars...>();
}

int main() {
  return "{}"_format(42).size();
}

[ldionne]

Not only could you perform the safety checks at compile-time, but you could also generate optimal code by parsing everything at compile-time.

[vitaut]

Parsing of format specs is now constexpr and works with GCC 6.3, AppleClang 7.3 and Visual C++ 2017. Now need to figure out how to report errors at compile time. Can't use static_assert because it requires message to be a string literal.

[Rush]

@vitaut tried to compile your std branch with "g++ (SUSE Linux) 7.2.1 20171020 [gcc-7-branch revision 253932]" in order to test constexpr but got a lot of errors. Is your CI catching those?

[vitaut]

@Rush, the current revision (d0f6d96) is broken at the moment. Please use dfdb1ad instead. I'll fix it shortly.

[vitaut]

@Rush, the std branch should be fixed now. Let me know if you still see errors on your version of gcc.

[vitaut]

As of f03a35a the format string checks are complete. Opened a separate issue for optimal code generation: #613.