Marshaling wchar_t on Linux uses wrong size

[lainy]

General

Tested on .NET Core 3.0.100-preview9-014004 on Arch Linux.

The following code should print out 4 on this system, but instead outputs 2.
Printing sizeof(wchar_t) from a C program gives 4.

Discovered when trying to interop with a C library that has a struct with wchar_t[5] inside.

using System;
using System.Runtime.InteropServices;

namespace dotnet_core_wchar_t_size
{
    class Program
    {
        [StructLayout(LayoutKind.Sequential, CharSet = CharSet.Unicode)]
        internal struct test
        {
            public char c;
        }

        static void Main(string[] args)
        {
            Console.WriteLine(Marshal.SizeOf(typeof(test)));
        }
    }
}

[tannergooding]

CharSet.Unicode is documented to be

Name	Value	Description
Unicode	3	Marshal strings as Unicode 2-byte character strings.

[jkoritzinsky]

More specifically, CharSet.Unicode is documented to map to a native char16_t on Unix-like platforms: https://docs.microsoft.com/dotnet/standard/native-interop/charset

[filmor]

@jkoritzinsky Is there a reasonable alternative, though? To have platform-independent bindings to an API using wchar_t, it would be usefull to have a CharSet.WCharT that is UTF16 on Windows and UTF32 essentially everywhere else. Apart from .NET tooling I have never seen an API that switches between UTF8 and UTF16 like CharSet.Auto does.