Might fix issue 58

[lemire]

Fixes #58

[lemire]

@gfoidl So hopefully you see what I had in mind. It still leaves open bug possibilities with big endian systems because we do not test. We are smart people but even smart people can't predict what could go wrong on systems they do not have access to.

[gfoidl]

Yep, that's fine. My mind was too knotty here...

In #49 care needs to be taken at

csFastFloat/csFastFloat/Utils/Utils.cs

Lines 38 to 42 in c108400

	unsafe internal static uint parse_eight_digits_unrolled(byte* chars)
	{
	ulong val = Unsafe.ReadUnaligned<ulong>(chars);
	return parse_eight_digits_unrolled(val);
	}

csFastFloat/csFastFloat/Utils/Utils.cs

Lines 55 to 59 in c108400

	unsafe internal static bool is_made_of_eight_digits_fast(byte* chars)
	{
	ulong val = Unsafe.ReadUnaligned<ulong>(chars);
	return is_made_of_eight_digits_fast(val);
	}

The Unsafe.ReadUnaligned is like a reinterpret_cast<ulong*> from an unsigned char source in C++ terms.

[lemire]

@gfoidl I am aware, but this is guarded... See...

    unsafe internal static bool is_made_of_eight_digits_fast(ulong val)
    {
      // We only enable paths depending on this function on little endian
      // platforms (it happens to be effectively nearly everywhere).
      return BitConverter.IsLittleEndian && (((val & 0xF0F0F0F0F0F0F0F0) |
               (((val + 0x0606060606060606) & 0xF0F0F0F0F0F0F0F0) >> 4)) ==
              0x3333333333333333);
    }

So is_made_of_eight_digits_fast will always return false on big endian systems.

[gfoidl]

Then it's perfect, simple and clean 👍

[lemire]

Merging with approval from @gfoidl