[PATCH] Implement "P0631R4 Math Constants" for C++20

[Jonathan Wakely]

On 31/07/19 21:16 +0200, Jakub Jelinek wrote:
>On Wed, Jul 31, 2019 at 07:58:55PM +0100, Jonathan Wakely wrote:
>> > Perhaps add __extension__ before the literals too?
>>
>> Does that do anything for Q literals? I thought I'd tried it
>> previously and decided it didn't. I'm happy to add it though.
>
>You're right, it works well that way only in C, not in C++, on say:
>long double x = 1.234Q;
>long double y = __extension__ 1.234Q;
>This is pedwarned not when actually parsing the number, but
>when tokenizing it, which for C++ we do before parsing for all tokens.
>
>We could add a hack for it, say don't emit the pedwarn if the previous token
>is RID_EXTENSION, but it wouldn't be anything close to how we handle
>__extension__ during parsing (where pedantic is disabled while parsing the
>whole cast expression after it).

Ah yes, I definitely looked into this before.

If I use the Q suffix like so ...

#if !defined(__STRICT_ANSI__) && defined(_GLIBCXX_USE_FLOAT128)
  template<>
    inline constexpr __float128 e_v<__float128>
      = 2.718281828459045235360287471352662498Q;

...
#endif

Then it produces 13 pedwarns with -pedantic -Wsystem-headers:

In file included from num.cc:1:
/home/jwakely/gcc/10/include/c++/10.0.0/numbers:139:7: warning: non-standard suffix on floating constant [-Wpedantic]
  139 |       = 2.718281828459045235360287471352662498Q;
      |       ^

Users would get those warnings even if not using the __float128
constants. Admittedly, if they use -pedantic combined with
-std=gnu++2a, and combine that with -Wsystem-headers, they deserve
what they get.

I think it's better to have accurate __float128 constants even if they
warn with certain combinations of warning flags.