Add Random Span-based APIs

[stephentoub]

Separated out of https://github.com/dotnet/corefx/issues/21281 for tracking purposes.

• Implement in System.Private.CoreLib in coreclr (corert shares the same file)
• Expose from System.Runtime.Extensions contract in corefx
• Add tests to System.Runtime.Extensions tests in corefx

namespace System
{
    public class Random
    {
        public virtual void NextBytes(Span<byte> buffer);
        …
    }
}

Implementation should be in terms of Next(). The NextBytes(byte[]) base method is implemented in terms of InternalSample, but NextBytes(Span<byte>) shouldn't be implemented in terms of that, as it's not public and thus an existing derived type couldn't have overridden it, and thus the new virtual wouldn't pick up existing behaviors. It also shouldn't be in terms of NextBytes(byte[]), as that would require allocating/copying an array, largely defeating the purpose of the method (at least for the base class). But since Next() on the base class just delegates to InternalSample(), Next() is a reasonable thing to base the implementation on, e.g.

public virtual void NextBytes(Span<byte> buffer)
{
    for (int i = 0; i < buffer.Length; i++)
    {
        buffer[i] = (byte)(Next() & 0xFF);
    }
}

[MaggieTsang]

@stephentoub Should the existing NextBytes(byte[]) also call Next()? Otherwise, if you derive the class NextBytes(Span<byte>) and NextBytes(byte[]) won't give identity results if you override Next().

[stephentoub]

We shouldn't change the existing NextBytes(byte[]), as that could be a breaking change (it will now be impacted by a virtual it wasn't previously impacted by), and it could have a performance impact (each byte would then incur an extra virtual method call instead of a non-virtual call).

[jnm2]

I'm curious, why is it written as (byte)(Next() & 0xFF) instead of (byte)Next()?

[stephentoub]

It could be 😄

[jnm2]

😄 I was curious what NextBytes(byte[]) was doing... now I wish I hadn't looked...

        public virtual void NextBytes(byte[] buffer)
        {
            if (buffer == null) throw new ArgumentNullException(nameof(buffer));
            Contract.EndContractBlock();
            for (int i = 0; i < buffer.Length; i++)
            {
                buffer[i] = (byte)(InternalSample() % (Byte.MaxValue + 1));
            }
        }

[stephentoub]

I basically just copied NextBytes for my example (changing the % 256 to & 0xFF).

But, yeah, that's unnecessarily complex. And because the JIT doesn't know that InternalSample() returns a non-negative value, the resulting asm is probably going to be worse than it needs to be. I think it's unlikely to actually matter though.

[stephentoub]

Fixed by dotnet/coreclr#13708 and dotnet/corefx#23692