Task.Run never completes when testing using a HttpClient

Question

Today I've run into a problem when creating a Web API using async ApiControllers. I'm using MongoDB and since the C# driver doesn't support async, I tried to implement it in my repository layer.

The resulting method in the Building repository looked like this:

public async Task<IEnumerable<Building>> GetAll()
{
    var tcs = new TaskCompletetionSource<IEnumerable<Building>>();

    await Task.Run(() => {
        var c = this.MongoDbCollection.FindAll();
        tcs.SetResult(c);
    });

    return await tcs.Task;
}

Now this works perfectly when testing the repository by itself using NUnit.

But when testing from the controller (using a HttpClient) it never proceeds to the "return" line after running tcs.SetResult(c). The test just keeps running until i abort it manually.

When I remove the Task.Run code and do everything synchronously then everything works as it should:

public async Task<IEnumerable<Building>> GetAll()
{
    var c = this.MongoDbCollection.FindAll();

    return c;
}

Does anyone have any idea why I experience different behaviors when testing the repository + database and when testing controller + repository + database?

The controller method looks like this: (buildingRepository is injected in the constructor using Ninject)

public async Task<HttpResponseMessage> Get()
{
    var result = await this.buildingRepository.GetAll();
    return Request.CreateResponse(HttpStatusCode.OK, result);
}

EDIT: Here are also the test methods. The first one is the one that's not working: (this.client is a HttpClient object with the accept-header set to "application/json")

[Test]
public void Get_WhenBuildingsExist_ShouldReturnBuilding()
{
    var task = this.client.GetAsync("/api/building/");

    var result = task.Result.Content.ReadAsStringAsync().Result;

    var o = JsonConvert.DeserializeObject<IEnumerable<Building>>(result);

    Assert.That(o.Any());
}

[Test]
public void Get_WhenBuildingsExist_ShouldReturnAtLeastOneBuilding()
{
    var buildings = this.buildingRepository.GetAll().Result;

    Assert.That(buildings.Any());
}

Answer 1

There is a post I read that explains why invoking .Results from an asynchronous task is a bad idea but I don't have it available at the moment. Basically, you're killing the async handling by doing this. Try changing your test as follow:

[Test]
public void Get_WhenBuildingsExist_ShouldReturnBuilding()
{
    var task = this.client.GetAsync("/api/building/");

    var resultTask = task.Result.Content.ReadAsStringAsync();
    resultTask.Wait();

    var result = resultTask.Result;
    var o = JsonConvert.DeserializeObject<IEnumerable<Building>>(result);

    Assert.That(o.Any());
}

Answer 2

The best option is to upgrade to a version of NUnit that supports async unit tests, and change all Result/Wait calls to await:

[Test]
public async Task Get_WhenBuildingsExist_ShouldReturnBuilding()
{
  var content = await this.client.GetAsync("/api/building/");
  var result = await content.ReadAsStringAsync();
  var o = JsonConvert.DeserializeObject<IEnumerable<Building>>(result);

  Assert.That(o.Any());
}

If this isn't possible (e.g., Xamarin still runs a very old version of NUnit as of this writing), then you can use AsyncContext from my AsyncEx library:

[Test]
public void Get_WhenBuildingsExist_ShouldReturnBuilding()
{
  var o = AsyncContext.Run(async () =>
  {
    var content = await this.client.GetAsync("/api/building/");
    var result = await content.ReadAsStringAsync();
    return JsonConvert.DeserializeObject<IEnumerable<Building>>(result);
  });

  Assert.That(o.Any());
}