EBNF incorrect: stackalloc_initializer can be empty (draft-v8).

[kaby76]

Describe the bug
Consider the following program.

using System;

unsafe class Program
{
    static void Main()
    {
        // Empty stackalloc initializer -- allocates a zero-length block.
        // Valid C# (accepted by Roslyn); the grammar rule for
        // stackalloc_initializer should allow an empty initializer list.
        int* p = stackalloc int[] {};
        Console.WriteLine("Empty stackalloc: length inferred as 0, pointer is non-null: " + ((IntPtr)p != IntPtr.Zero));

        // Non-empty for contrast
        int* q = stackalloc int[] { 1, 2, 3 };
        Console.WriteLine("Non-empty stackalloc[0]: " + q[0]);
    }
}

This code compiles and runs fine in version 8 of the C# Roslyn compiler. In fact, it explicitly tests this for version 8 of the C# language.

foo.zip

https://github.com/dotnet/roslyn/blob/2ad4aabc7a9dada097e54e544ebba48ab1c05074/src/Compilers/CSharp/Test/Emit/CodeGen/CodeGenStackAllocInitializerTests.cs#L369

However, the draft-v8 spec is wrong:

csharpstandard/standard/expressions.md

Lines 3411 to 3417 in 089fae9

	stackalloc_initializer
	: '{' stackalloc_initializer_element_list '}'
	;
	stackalloc_initializer_element_list
	: stackalloc_element_initializer (',' stackalloc_element_initializer)* ','?
	;

The rule should be:

stackalloc_initializer
     : '{' stackalloc_initializer_element_list? '}'
     ;

Or, stackalloc_initializer_element_list should allow for empty.

This feature was introduced in C# 7, so that EBNF is probably wrong as well.