base/mpi.h: also instantiate for signed long long int

[tamiko]

Vaguely in reference to #16493

[tamiko]

/rebuild

[masterleinad]

What about MPI_DEVICE_SCALARS for consistency?

[tamiko]

@masterleinad Good point. Updated.

[bangerth]

Out of curiosity, it is true that on all systems we work on, two out of int, long int, and long long int are the same. Does the compiler not complain in those cases?

Should we instead instantiate for int and int64_t?

[tamiko]

@bangerth I think according to the standard int, long int and long long int are distinct types, meaning:

template <typename C> void foo(){};
template void foo<int>();
template void foo<long int>();  // ok
template void foo<long long int>(); // ok
template void foo<std::int64_t>(); // error: duplicate instantiation

[drwells]

It would be deeply weird if

static_assert(std::is_same_v<long int, long long int>);

succeeded on some platforms but failed on others.

[bangerth]

Not weird at all. std::is_same<int,std::int16_t> is also true on some platforms and false on others. The length of integers has always been platform dependent. On some platforms, long int is 32 bit and on others it is 64 bit. long long int is pretty generally 64 bit, but I don't think the standard guarantees that it is.

[tamiko]

The difference is that std::intN_t is allowed to be a typedef whereas int, long, and long long must be different types. Even if they are functionally the same.

[drwells]

It's an interesting question and I had to look it up to be sure. In 17.4.2 all the cstdint types (int16_t et al) are declared with using, whereas in 6.8.2 signed char, short int, etc are "the five standard signed integer types".