Proper way to implement methods that return Task<T>

Question

For simplicity let's imagine we have a method that should return an object while doing some heavy operation. There're two ways to implement:

public Task<object> Foo()
{
    return Task.Run(() =>
    {
        // some heavy synchronous stuff.

        return new object();
    }
}

And

public async Task<object> Foo()
{
    return await Task.Run(() =>
    {
        // some heavy stuff
        return new object();
    }
}

After examining generated IL there're two completely different things generated:

.method public hidebysig 
    instance class [mscorlib]System.Threading.Tasks.Task`1<object> Foo () cil managed 
{
    // Method begins at RVA 0x2050
    // Code size 42 (0x2a)
    .maxstack 2
    .locals init (
        [0] class [mscorlib]System.Threading.Tasks.Task`1<object>
    )

    IL_0000: nop
    IL_0001: ldsfld class [mscorlib]System.Func`1<object> AsyncTest.Class1/'<>c'::'<>9__0_0'
    IL_0006: dup
    IL_0007: brtrue.s IL_0020

    IL_0009: pop
    IL_000a: ldsfld class AsyncTest.Class1/'<>c' AsyncTest.Class1/'<>c'::'<>9'
    IL_000f: ldftn instance object AsyncTest.Class1/'<>c'::'<Foo>b__0_0'()
    IL_0015: newobj instance void class [mscorlib]System.Func`1<object>::.ctor(object, native int)
    IL_001a: dup
    IL_001b: stsfld class [mscorlib]System.Func`1<object> AsyncTest.Class1/'<>c'::'<>9__0_0'

    IL_0020: call class [mscorlib]System.Threading.Tasks.Task`1<!!0> [mscorlib]System.Threading.Tasks.Task::Run<object>(class [mscorlib]System.Func`1<!!0>)
    IL_0025: stloc.0
    IL_0026: br.s IL_0028

    IL_0028: ldloc.0
    IL_0029: ret
}

And

.method public hidebysig 
    instance class [mscorlib]System.Threading.Tasks.Task`1<object> Foo () cil managed 
{
    .custom instance void [mscorlib]System.Runtime.CompilerServices.AsyncStateMachineAttribute::.ctor(class [mscorlib]System.Type) = (
        01 00 1a 41 73 79 6e 63 54 65 73 74 2e 43 6c 61
        73 73 31 2b 3c 42 61 72 3e 64 5f 5f 31 00 00
    )
    .custom instance void [mscorlib]System.Diagnostics.DebuggerStepThroughAttribute::.ctor() = (
        01 00 00 00
    )
    // Method begins at RVA 0x2088
    // Code size 59 (0x3b)
    .maxstack 2
    .locals init (
        [0] class AsyncTest.Class1/'<Foo>d__1',
        [1] valuetype [mscorlib]System.Runtime.CompilerServices.AsyncTaskMethodBuilder`1<object>
    )

    IL_0000: newobj instance void AsyncTest.Class1/'<Foo>d__1'::.ctor()
    IL_0005: stloc.0
    IL_0006: ldloc.0
    IL_0007: ldarg.0
    IL_0008: stfld class AsyncTest.Class1 AsyncTest.Class1/'<Foo>d__1'::'<>4__this'
    IL_000d: ldloc.0
    IL_000e: call valuetype [mscorlib]System.Runtime.CompilerServices.AsyncTaskMethodBuilder`1<!0> valuetype [mscorlib]System.Runtime.CompilerServices.AsyncTaskMethodBuilder`1<object>::Create()
    IL_0013: stfld valuetype [mscorlib]System.Runtime.CompilerServices.AsyncTaskMethodBuilder`1<object> AsyncTest.Class1/'<Foo>d__1'::'<>t__builder'
    IL_0018: ldloc.0
    IL_0019: ldc.i4.m1
    IL_001a: stfld int32 AsyncTest.Class1/'<Foo>d__1'::'<>1__state'
    IL_001f: ldloc.0
    IL_0020: ldfld valuetype [mscorlib]System.Runtime.CompilerServices.AsyncTaskMethodBuilder`1<object> AsyncTest.Class1/'<Foo>d__1'::'<>t__builder'
    IL_0025: stloc.1
    IL_0026: ldloca.s 1
    IL_0028: ldloca.s 0
    IL_002a: call instance void valuetype [mscorlib]System.Runtime.CompilerServices.AsyncTaskMethodBuilder`1<object>::Start<class AsyncTest.Class1/'<Foo>d__1'>(!!0&)
    IL_002f: ldloc.0
    IL_0030: ldflda valuetype [mscorlib]System.Runtime.CompilerServices.AsyncTaskMethodBuilder`1<object> AsyncTest.Class1/'<Foo>d__1'::'<>t__builder'
    IL_0035: call instance class [mscorlib]System.Threading.Tasks.Task`1<!0> valuetype [mscorlib]System.Runtime.CompilerServices.AsyncTaskMethodBuilder`1<object>::get_Task()
    IL_003a: ret
}

As you can see in the first case logic is straightforward, lambda function is created and then call to the Task.Run is generated and the result is returned. In the second example instance of AsyncTaskMethodBuilder is created and then task is actually built and returned. Since I always expected foo method to be called as await Foo() at some higher level, I have always used the first example. However, I see the latter more often. So which approach is correct? What pros and cons do each have?

---

Real world example

Let's say we have UserStore which has method Task<User> GetUserByNameAsync(string userName) which is used inside web api controller like:

public async Task<IHttpActionResult> FindUser(string userName)
{
    var user = await _userStore.GetUserByNameAsync(userName);

    if (user == null)
    {
        return NotFound();
    }

    return Ok(user);
}

Which implementation of Task<User> GetUserByNameAsync(string userName) would be correct?

public Task<User> GetUserByNameAsync(string userName)
{
    return _dbContext.Users.FirstOrDefaultAsync(user => user.UserName == userName);
}

or

public async Task<User> GetUserNameAsync(string userName)
{
    return await _dbContext.Users.FirstOrDefaultAsync(user => user.UserName == username);
}

Answer 1

So which approach is correct?

Neither.

If you have synchronous work to do, then the API should be synchronous:

public object Foo()
{
    // some heavy synchronous stuff.

    return new object();
}

If the calling method can block its thread (i.e., it's an ASP.NET call, or it's running on a thread pool thread), then it just calls it directly:

var result = Foo();

And if the calling thread can't block it's thread (i.e., it's running on the UI thread), then it can run Foo on the thread pool:

var result = await Task.Run(() => Foo());

As I describe on my blog, Task.Run should be used for invocation, not implementation.

---

Real World Example

(which is a completely different scenario)

Which implementation of Task GetUserByNameAsync(string userName) would be correct?

Either one is acceptable. The one with async and await has some extra overhead, but it won't be noticeable at runtime (assuming the thing you're awaiting actually does I/O, which is true in the general case).

Note that if there's other code in the method, then the one with async and await is better. This is a common mistake:

Task<string> MyFuncAsync()
{
  using (var client = new HttpClient())
    return client.GetStringAsync("http://www.example.com/");
}

In this case, the HttpClient is disposed before the task completes.

Another thing to note is that exceptions before returning the task are thrown differently:

Task<string> MyFuncAsync(int id)
{
  ... // Something that throws InvalidOperationException
  return OtherFuncAsync();
}

Since there is no async, the exception is not placed on the returned task; it is thrown directly. This can confuse the calling code if it does something more complex than just awaiting the task:

var task1 = MyFuncAsync(1); // Exception is thrown here.
var task2 = MyFuncAsync(2);
...
try
{
  await Task.WhenAll(task1, task2);
}
catch (InvalidOperationException)
{
  // Exception is not caught here. It was thrown at the first line.
}