Error `Async test method must have non-void return type` when upgrading from NUnit 2 to NUnit 3

Question

I have to refactor am unit test from NUNIT 2 to NUNIT 3 and the following syntax throws an error:

var expectedResponseMessage = new HttpResponseMessage();
Func<Task<HttpResponseMessage>> continuation = 
   () => Task.Factory.StartNew(() => expectedResponseMessage);

Error:

Async test method must have non-void return type

How may I rewrite this? I have tried many syntaxes but no luck.

Answer 1

The error

Async test method must have non-void return type

means that in NUnit 3+, an async Unit Test itself may not have a void return type (i.e. the method decorated with [Test] / [TestCase] etc). Instead, you can return an empty Task (Correct way in newer versions of NUnit with async test support):

[Test]
public async Task EnsureFoo()
{
     // Arrange

     // Act
     var myResult = await classBeingTested.DoSomethingAsync();

     // Assert
     Assert.IsNotNull(myResult);
     ...
}

In NUnit 2.x, this wasn't checked, so a async void unit test could slip into your unit test code base, i.e. of the form (Bad, don't do this)

[Test]
public async void Foo() // < -- Error : Async test method must have non-void return type
{
     var myResult = await classBeingTested.DoSomethingAsync();
     // Post continuation assertions here.
}

this is rather dangerous - the test doesn't return a Task which can be awaited by the runner^*, and could return before any continuations completed - e.g. any failures in the Asserts done in the continuation might not be reported.

Re : Your fake Task

Scheduling a Task just to return a fake response seems overkill, i.e. in most tests you should be able to use Task.FromResult to replace:

Func<Task<HttpResponseMessage>> continuation = 
   () => Task.Factory.StartNew(() => expectedResponseMessage);

With the cheaper:

Func<Task<HttpResponseMessage>> continuation = 
   () => Task.FromResult(expectedResponseMessage);

Task.FromResult returns an already completed task with the given return value - in most cases, this should suffice for your unit testing purposes, unless you really do want an independent Task to be executed on the threadpool.

---

_{* Actually, seemingly even earlier versions such as NUnit 2.6.4 had already identified the issue with async void tests, and incorporated a workaround}