Append Buffer Async Readback broken

gistfile1.txt

Compute Shader:

// Each #kernel tells which function to compile; you can have many kernels
#pragma kernel CSMain

// Create a RenderTexture with enableRandomWrite flag and set it
// with cs.SetTexture
AppendStructuredBuffer<float> Result;

[numthreads(256,1,1)]
void CSMain (uint3 id : SV_DispatchThreadID)
{
    if (id.x < 3000)
    {
        Result.Append(1);
    }
}


Script:

using System;
using System.Collections;
using System.Collections.Generic;
using Unity.Collections;
using UnityEngine;
using UnityEngine.Rendering;

[ExecuteAlways]
public class TestAppendBuffer : MonoBehaviour
{
    public enum Mode
    {
        MonoAllocating,
        AsyncReadback,
        AsyncReadbackBlocking
    }

    public Mode mode = Mode.MonoAllocating;

    NativeArray<float> asArray;
    GraphicsBuffer outputBuf;
    private void OnEnable()
    {
        int numToDo = 3000;
        outputBuf = new GraphicsBuffer(GraphicsBuffer.Target.Append, numToDo, 4);
        outputBuf.SetCounterValue(0);

        ComputeShader cs = Resources.Load<ComputeShader>("TestAppendBuffer");
        int kernel = cs.FindKernel("CSMain");
        cs.SetBuffer(kernel, "Result", outputBuf);
        cs.Dispatch(kernel, Mathf.CeilToInt(numToDo / 256.0f), 1, 1);

        if (mode == Mode.MonoAllocating)
        {
            var data = new float[3000];
            outputBuf.GetData(data, 0, 0, 3000);
            int valid = 0;
            int invalid = 0;
            foreach (var m in data)
            {
                if (m == 0)
                {
                    invalid++;
                }
                else
                    valid++;
            }
            if (valid == 3000)
            {
                Debug.Log("Working Correctly, all 3000 entries processed");
            }
            else
            {
                Debug.Log("Failed, " + valid + " entries processed, " + invalid + " not processed");
            }
            outputBuf.Dispose();
            return;
        }
        else if (mode == Mode.AsyncReadback)
        {
            asArray = new NativeArray<float>(numToDo, Allocator.Persistent);
            var req = AsyncGPUReadback.RequestIntoNativeArray(ref asArray, outputBuf, numToDo, 0);
            req.WaitForCompletion();
            int valid = 0;
            int invalid = 0;
            foreach (var m in asArray)
            {
                if (m == 0)
                {
                    invalid++;
                }
                else
                    valid++;
            }
            if (valid == 3000)
            {
                Debug.Log("Working Correctly, all 3000 entries processed");
            }
            else
            {
                Debug.Log("Failed, " + valid + " entries processed, " + invalid + " not processed");
            }
            outputBuf.Dispose();
            asArray.Dispose();
        }
        else
        {
            asArray = new NativeArray<float>(numToDo, Allocator.Persistent);
            var req = AsyncGPUReadback.RequestIntoNativeArray(ref asArray, outputBuf, numToDo, 0, callback);
        }
     


    }

    private void callback(AsyncGPUReadbackRequest request)
    {
        var data = request.GetData<float>();
        int valid = 0;
        int invalid = 0;
        foreach (var m in data)
        {
            if (m == 0)
            {
                invalid++;
            }
            else
                valid++;
        }

        if (valid == 3000)
        {
            Debug.Log("Working Correctly, all 3000 entries processed");
        }
        else
        {
            Debug.Log("Failed, " + valid + " entries processed, " + invalid + " not processed");
        }
        outputBuf.Dispose();
        asArray.Dispose();
    }
}

To repro: Add the component and put the compute shader in a Resource folder. Toggle enable on/off to test. Change modes to try different modes, all the aync/non-allocating versions are broken.