Currently the result of nextafter(x, y) against NaN is undefined at least in the document. Actually, although nextafter(NaN, y) returns NaN as a result of nextDown(NaN), nextafter(x, NaN) returns nextDown(x), as below:

void main()
{
    import std;

    auto a = 4.0;
    writefln("%a", a); // 0x1p+2
    writefln("%a", nextafter(a.nan, a)); // nan, it's fine.
    writefln("%a", nextafter(a, a.nan)); // 0x1.fffffffffffffp+1, but should be nan.
}

On POSIX, the man page of nextafter(3) describes ambiguously:
> If x or y is NaN, a NaN shall be returned.
http://man7.org/linux/man-pages/man3/nextafter.3p.html

On the other hand, the Linux version says:
> If x or y is a NaN, a NaN is returned.
http://man7.org/linux/man-pages/man3/nextafter.3.html

Similarly, Microsoft says:
> If either x or y is a NAN, then the return value is one of the input NANs.
https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/nextafter-functions?view=vs-2019