Should an Async API Ever Throw Synchronously

Should an async API ever throw synchronously?

Ultimately the decision to synchronously throw or not is up to you, and you will likely find people who argue either side. The important thing is to document the behavior and maintain consistency in the behavior.

My opinion on the matter is that your second option - passing the error into the callback - seems more elegant. Otherwise you end up with code that looks like this:

try {
    getUserById(7, function (response) {
       if (response.isSuccess) {
           //Success case
       } else {
           //Failure case
       }
    });
} catch (error) {
    //Other failure case
}

The control flow here is slightly confusing.

It seems like it would be better to have a single if / else if / else structure in the callback and forgo the surrounding try / catch.

Why do I need to await an async function when it is not supposedly returning a Promise?

All async functions return a promise. All of them.

That promise will eventually resolve with whatever value you return from the async function.

await only blocks execution internal to the async function. It does not block anything outside of the function. Conceptually, an async function starts to execute and as soon as it hits an await instruction, it immediately returns an unfulfilled promise (in the pending state) from the function and the outside execution world gets that promise and continues to execute.

Sometime later, the internal promise that was being awaited will resolve and then the execution of the rest of the internals of the function will continue. Eventually the internals of the function will finish and return a value. That will trigger resolving the promise that was returned from the function with that return value.

FYI, there's a lot of superfluous stuff in your load() function. You can change it from this:

async function load() {
    const data = await new Promise(resolve => {
        setTimeout(() => resolve([1, 2, 3]), 10);
    }).then(data => data.map(i => i * 10));
    console.log(`Data inside the function: ${JSON.stringify(data)}`);
    return data;
}

to this:

function load() {
    return new Promise(resolve => {
        setTimeout(() => resolve([1, 2, 3]), 10);
    }).then(data => data.map(i => i * 10));
}

Then, use it like this:

load().then(result => {
    console.log(result);
});

Or, I prefer to encapsulate the manual creation of promise in their own function like this:

function delay(t, v) {
    return new Promise(resolve => {
        setTimeout(resolve.bind(null, v), t);
    });
}

function load() {
    return delay(10, [1, 2, 3]).then(data => data.map(i => i * 10));
}

And, it turns out this little delay() function is generally useful in lots of places where you want to delay a promise chain.

Thanks to everyone participating and providing me with insight. But I'm still confused how should I be using await and async.

First off, most of the time you only mark a function async if you need to use await inside the function.

Second, you most commonly use await (from within an async function) when you have multiple asynchronous operations and you want to sequence them - often because the first one provides a result that is used as input to the second. You can use await when all you have is a single asynchronous operation, but it doesn't really offer much of an advantage over a simple .then().

Here are a few examples of good reasons to use async/await:

Sequencing multiple asynchronous operations

Imagine you have getFromDatabase(), getTheUrl() and getTheContent() that are all asynchronous. If any fails, you would want to just reject the returned promise with the first error.

Here's how this looks without async/await:

function run() {
    return getFromDatabase(someArg).then(key => {
        return getTheURL(key);
    }).then(url => {
        return getTheContent(url);
    }).then(content => {
        // some final processing
        return finalValue;
    });
}

Here's how this looks with async/await:

async function run(someArg) {
    let key = await getFromDatabase(someArg);
    let url = await getTheURL(key);
    let content = await getTheContent(url);
    // some final processing
    return finalValue;  
}

In both cases, the function returns a promise that resolves with the finalValue so these two implementations are used the same by the caller:

run(someArg).then(finalValue => {
    console.log(finalValue);
}).catch(err => {
    console.log(err);
});

But, you will notice that the async/await implementation has more of a serialized, synchronous look to it and looks more like non-asynchronous code. Many find this easier to write, easier to read and easier to maintain. The more processing you have between steps, including branching, the more advantages accrue to the async/await version.

Automatically catching both rejected promises and synchronous exceptions

As I said earlier, async functions always return a promise. They also have to built-in error handling that automatically propagates errors back to that returned promise.

It goes without saying that if you manually return a promise from the async function and that promise rejects, then the promise returned from the async function will reject.

But also, if you are using await and any promise you are awaiting rejects and you don't have a .catch() on the promise and don't have a try/catch around it, then the promise the function returns will automatically reject. So, back in our previous example of this:

async function run(someArg) {
    let key = await getFromDatabase(someArg);
    let url = await getTheURL(key);
    let content = await getTheContent(url);
    // some final processing
    return finalValue;
}

If any of the three promises that are being awaited rejects, then the function will short circuit (stop executing any more code in the function) and reject the async returned promise. So, you get this form of error handling for free.

Then lastly, an async function also catches synchronous exceptions for you and turns them into a rejected promise.

In a normal function that returns a promise such as we had earlier:

function run() {
    return getFromDatabase(someArg).then(key => {
        return getTheURL(key);
    }).then(url => {
        return getTheContent(url);
    }).then(content => {
        // some final processing
        return finalValue;
    });
}

If getFromDatabase() throws a synchronous exception (perhaps triggered because someArg is invalid), then this overall function run() will throw synchronously. That means that for the caller to catch all possible errors from run(), they have to both surround it with a try/catch to catch the synchronous exceptions and use a .catch() to catch the rejected promise:

try {
    run(someArg).then(finalValue => {
        console.log(finalValue);
    }).catch(err => {
        console.log(err);
    });
} catch(e) {
    console.log(err);
}

This is messy and a bit repetitive. But, when run() is declared async, then it will NEVER throw synchronously because any synchronous exception is automatically converted to a rejected promise so you can be sure you are capturing all possible errors when it's written this way:

async function run(someArg) {
    let key = await getFromDatabase(someArg);
    let url = await getTheURL(key);
    let content = await getTheContent(url);
    // some final processing
    return finalValue;
}

// will catch all possible errors from run()
run(someArg).then(finalValue => {
    console.log(finalValue);
}).catch(err => {
    console.log(err);
});

Should I always call all of my function with an await?

First, you would only ever use await with a function that returns a promise as await offers no usefulness if the function does not return a promise (just adding to the clutter of your code if not needed).

Second, whether you use await or not depends upon the context of both the calling function (since you HAVE to be in an async function to use await and on the flow of logic and whether it benefits from using await or not.

Places where it's pointless to use await

async function getKey(someArg) {
    let key = await getFromDatabase(someArg);
    return key;
}

The await here isn't doing anything useful. You're not sequencing multiple async operations and you're not doing any processing on the return value. You can accomplish the exact same code by just returning the promise directly:

async function getKey(someArg) {
    return getFromDatabase(someArg);
}

And, if you know that getFromDatabase() never throws synchronously, you can even remove the async from the declaration:

function getKey(someArg) {
    return getFromDatabase(someArg);
}

Let's say I'm writing a code composed of multiple functions within multiple files. If I end up using a library which returns a Promise or it's an async function, should I trace back all my function calls from the asynchronous point to the entry point of the application and add an await before all the function calls after making them async?

This is a bit too general of an ask that's hard to answer in a general case. Here are some thoughts along this general direction:

Once any part of your result that you're trying to return from your function A() is asynchronous or uses any asynchronous operation to obtain, the function itself is asynchronous. In plain Javascript, you can never return an asynchronous result synchronously so your function must use an asynchronous method to return the result (promise, callback, event, etc...).
Any function B() that calls your asynchronous function A() that is also trying to return a result based on what it gets from A() is now also asynchronous and also must communicate its result back using an asynchronous mechanism. This is true for a function C() that calls B() and needs to communicate back its result to the caller. So, you can say that asynchronous behavior is infectious. Until you reach some point in the call chain where you no longer need to communicate back a result, everything has to use asynchronous mechanisms to communicate the result, error and completion.
There's no specific need to mark a function async unless you specifically need one of the benefits of an async function such as the ability to use await inside that function or the automatic error handling it provides. You can write functions that returning promises just fine without using async on the function declaration. So, "NO" I don't go back up the call chain making everything async. I only make a function async if there's a specific reason to do so. Usually that reason is that I want to use await inside the function, but there is also the automatic catching of synchronous exceptions that get turned into promise rejections that I described earlier. You would not generally need that with well behaved code, but it is sometimes useful with poorly behaved code orcode with an undefined behavior.
await is also only used when there's a specific reason for it. I don't just automatically use it on every function that returns a promise. I've described above reasons to use it. One can still use .then() just fine for processing the result from a single function call that returns a promise. In some cases, it's just a matter of personal style whether you want to use .then() or await and there is no particular reason it has to be one one way or the other.

Or maybe I should just get into the habit of calling all my functions with an await regardless of whether they are async or not?

Absolutely NOT! First off, the last thing you want to do is to take perfectly synchronous code and unnecessarily make it asynchronous or even make it look asynchronous. asynchronous code (even with async and await) is more complicated to write, debug, understand and maintain than synchronous code so you would never want to unnecessarily make synchronous code into asynchronous code by adding async/await into it:

For example, you would never do this:

async function random(min, max) {
    let r = await Math.random();
    return Math.floor((r * (max - min)) + min);
}

First off, this is a perfectly synchronous operation that can be coded like this:

function random(min, max) {
    let r = Math.random();
    return Math.floor((r * (max - min)) + min);
}

Second off, that first async implementation has made the function a lot hard to use as it now has an asynchronous result:

random(1,10).then(r => {
    console.log(r);
});

Instead of just the simple synchronous use:

console.log(random(1,10));

Should async function never ever throw?

Your assumptions on the async code are correct. See this post by Isaac Schlueter himself on the topic:

The pattern in node is that sync methods throw, and async methods pass
the error as the first argument to the callback. If the first
argument to your callback is falsey (usually null or undefined), then
all is well with the world.

http://groups.google.com/forum/#!msg/nodejs/W9UVJCKcJ7Q/rzseRbourCUJ

Trying to call Async method synchronously. It waits on Task.Result forever

This is a classic deadlock. The UI is waiting on the async method to finish, but the async method trys to update the UI thread and BOOM, deadlock.

Curiously though, if I copy this pattern more or less verbatim into a
Console app, it works.

That's because your WinForm application has a custom SynchronizationContext. It is caputred implicitly and its job is to marshal work back onto the UI thread once returning from your await.

Should you really expose synchronous wrappers around asynchronous operations?, the answer is no.

There is a way out of it, but i dont really like it. If you absolutely have to (you don't) call your code synchronously (again, you really shouldn't), use ConfigureAwait(false) inside the async method. This instructs the awaitable not to capture the current synccontext, so it wont marshal work back onto the UI thread:

public async Task<string> MyMethodAsync() 
{ 
    using (var cl = new HttpClient()) 
    {
        return await cl.GetStringAsync("http://www.google.co.uk/")
                       .ConfigureAwait(false);
    }
}

Note that if you do this and then try to call any UI element afterwards, you'll end up with an InvalidOperationException since you won't be on the UI thread.

Initializing the UI via a constructor is a common pattern. Stephan Cleary has a very nice series on async which you can find here.

What am I doing wrong? Is this simply a bad pattern, and if so, what
pattern should I be using instead?

Yes, absolutely. If you want to expose both asynchronous and synchronous APIs, use the proper api's which wouldn't get you into this situation (a deadlock) in the firstcase. For example, if you want to expose a synchronous DownloadString, use WebClient instead.

How to call asynchronous method from synchronous method in C#?

Asynchronous programming does "grow" through the code base. It has been compared to a zombie virus. The best solution is to allow it to grow, but sometimes that's not possible.

I have written a few types in my Nito.AsyncEx library for dealing with a partially-asynchronous code base. There's no solution that works in every situation, though.

Solution A

If you have a simple asynchronous method that doesn't need to synchronize back to its context, then you can use Task.WaitAndUnwrapException:

var task = MyAsyncMethod();
var result = task.WaitAndUnwrapException();

You do not want to use Task.Wait or Task.Result because they wrap exceptions in AggregateException.

This solution is only appropriate if MyAsyncMethod does not synchronize back to its context. In other words, every await in MyAsyncMethod should end with ConfigureAwait(false). This means it can't update any UI elements or access the ASP.NET request context.

Solution B

If MyAsyncMethod does need to synchronize back to its context, then you may be able to use AsyncContext.RunTask to provide a nested context:

var result = AsyncContext.RunTask(MyAsyncMethod).Result;

*Update 4/14/2014: In more recent versions of the library the API is as follows:

var result = AsyncContext.Run(MyAsyncMethod);

(It's OK to use Task.Result in this example because RunTask will propagate Task exceptions).

The reason you may need AsyncContext.RunTask instead of Task.WaitAndUnwrapException is because of a rather subtle deadlock possibility that happens on WinForms/WPF/SL/ASP.NET:

A synchronous method calls an async method, obtaining a Task.
The synchronous method does a blocking wait on the Task.
The async method uses await without ConfigureAwait.
The Task cannot complete in this situation because it only completes when the async method is finished; the async method cannot complete because it is attempting to schedule its continuation to the SynchronizationContext, and WinForms/WPF/SL/ASP.NET will not allow the continuation to run because the synchronous method is already running in that context.

This is one reason why it's a good idea to use ConfigureAwait(false) within every async method as much as possible.

Solution C

AsyncContext.RunTask won't work in every scenario. For example, if the async method awaits something that requires a UI event to complete, then you'll deadlock even with the nested context. In that case, you could start the async method on the thread pool:

var task = Task.Run(async () => await MyAsyncMethod());
var result = task.WaitAndUnwrapException();

However, this solution requires a MyAsyncMethod that will work in the thread pool context. So it can't update UI elements or access the ASP.NET request context. And in that case, you may as well add ConfigureAwait(false) to its await statements, and use solution A.

Update, 2019-05-01: The current "least-worst practices" are in an MSDN article here.

Async and await to synchronous method using Task

You can run asynchronous code in a synchronous manner by using the following extensions.

https://stackoverflow.com/a/5097066/5062791

public static class AsyncHelpers
{
    /// <summary>
    /// Execute's an async Task<T> method which has a void return value synchronously
    /// </summary>
    /// <param name="task">Task<T> method to execute</param>
    public static void RunSync(Func<Task> task)
    {
        var oldContext = SynchronizationContext.Current;
        var synch = new ExclusiveSynchronizationContext();
        SynchronizationContext.SetSynchronizationContext(synch);
        synch.Post(async _ =>
        {
            try
            {
                await task();
            }
            catch (Exception e)
            {
                synch.InnerException = e;
                throw;
            }
            finally
            {
                synch.EndMessageLoop();
            }
        }, null);
        synch.BeginMessageLoop();

        SynchronizationContext.SetSynchronizationContext(oldContext);
    }

    /// <summary>
    /// Execute's an async Task<T> method which has a T return type synchronously
    /// </summary>
    /// <typeparam name="T">Return Type</typeparam>
    /// <param name="task">Task<T> method to execute</param>
    /// <returns></returns>
    public static T RunSync<T>(Func<Task<T>> task)
    {
        var oldContext = SynchronizationContext.Current;
        var synch = new ExclusiveSynchronizationContext();
        SynchronizationContext.SetSynchronizationContext(synch);
        T ret = default(T);
        synch.Post(async _ =>
        {
            try
            {
                ret = await task();
            }
            catch (Exception e)
            {
                synch.InnerException = e;
                throw;
            }
            finally
            {
                synch.EndMessageLoop();
            }
        }, null);
        synch.BeginMessageLoop();
        SynchronizationContext.SetSynchronizationContext(oldContext);
        return ret;
    }

    private class ExclusiveSynchronizationContext : SynchronizationContext
    {
        private bool done;
        public Exception InnerException { get; set; }
        readonly AutoResetEvent workItemsWaiting = new AutoResetEvent(false);
        readonly Queue<Tuple<SendOrPostCallback, object>> items =
            new Queue<Tuple<SendOrPostCallback, object>>();

        public override void Send(SendOrPostCallback d, object state)
        {
            throw new NotSupportedException("We cannot send to our same thread");
        }

        public override void Post(SendOrPostCallback d, object state)
        {
            lock (items)
            {
                items.Enqueue(Tuple.Create(d, state));
            }
            workItemsWaiting.Set();
        }

        public void EndMessageLoop()
        {
            Post(_ => done = true, null);
        }

        public void BeginMessageLoop()
        {
            while (!done)
            {
                Tuple<SendOrPostCallback, object> task = null;
                lock (items)
                {
                    if (items.Count > 0)
                    {
                        task = items.Dequeue();
                    }
                }
                if (task != null)
                {
                    task.Item1(task.Item2);
                    if (InnerException != null) // the method threw an exeption
                    {
                        throw new AggregateException("AsyncHelpers.Run method threw an exception.", InnerException);
                    }
                }
                else
                {
                    workItemsWaiting.WaitOne();
                }
            }
        }

        public override SynchronizationContext CreateCopy()
        {
            return this;
        }
    }
}

How would I run an async TaskT method synchronously?

Here's a workaround I found that works for all cases (including suspended dispatchers). It's not my code and I'm still working to fully understand it, but it does work.

It can be called using:

customerList = AsyncHelpers.RunSync<List<Customer>>(() => GetCustomers());

Code is from here

public static class AsyncHelpers
{
    /// <summary>
    /// Execute's an async Task<T> method which has a void return value synchronously
    /// </summary>
    /// <param name="task">Task<T> method to execute</param>
    public static void RunSync(Func<Task> task)
    {
        var oldContext = SynchronizationContext.Current;
        var synch = new ExclusiveSynchronizationContext();
        SynchronizationContext.SetSynchronizationContext(synch);
        synch.Post(async _ =>
        {
            try
            {
                await task();
            }
            catch (Exception e)
            {
                synch.InnerException = e;
                throw;
            }
            finally
            {
                synch.EndMessageLoop();
            }
        }, null);
        synch.BeginMessageLoop();

        SynchronizationContext.SetSynchronizationContext(oldContext);
    }

    /// <summary>
    /// Execute's an async Task<T> method which has a T return type synchronously
    /// </summary>
    /// <typeparam name="T">Return Type</typeparam>
    /// <param name="task">Task<T> method to execute</param>
    /// <returns></returns>
    public static T RunSync<T>(Func<Task<T>> task)
    {
        var oldContext = SynchronizationContext.Current;
        var synch = new ExclusiveSynchronizationContext();
        SynchronizationContext.SetSynchronizationContext(synch);
        T ret = default(T);
        synch.Post(async _ =>
        {
            try
            {
                ret = await task();
            }
            catch (Exception e)
            {
                synch.InnerException = e;
                throw;
            }
            finally
            {
                synch.EndMessageLoop();
            }
        }, null);
        synch.BeginMessageLoop();
        SynchronizationContext.SetSynchronizationContext(oldContext);
        return ret;
    }

    private class ExclusiveSynchronizationContext : SynchronizationContext
    {
        private bool done;
        public Exception InnerException { get; set; }
        readonly AutoResetEvent workItemsWaiting = new AutoResetEvent(false);
        readonly Queue<Tuple<SendOrPostCallback, object>> items =
            new Queue<Tuple<SendOrPostCallback, object>>();

        public override void Send(SendOrPostCallback d, object state)
        {
            throw new NotSupportedException("We cannot send to our same thread");
        }

        public override void Post(SendOrPostCallback d, object state)
        {
            lock (items)
            {
                items.Enqueue(Tuple.Create(d, state));
            }
            workItemsWaiting.Set();
        }

        public void EndMessageLoop()
        {
            Post(_ => done = true, null);
        }

        public void BeginMessageLoop()
        {
            while (!done)
            {
                Tuple<SendOrPostCallback, object> task = null;
                lock (items)
                {
                    if (items.Count > 0)
                    {
                        task = items.Dequeue();
                    }
                }
                if (task != null)
                {
                    task.Item1(task.Item2);
                    if (InnerException != null) // the method threw an exeption
                    {
                        throw new AggregateException("AsyncHelpers.Run method threw an exception.", InnerException);
                    }
                }
                else
                {
                    workItemsWaiting.WaitOne();
                }
            }
        }

        public override SynchronizationContext CreateCopy()
        {
            return this;
        }
    }
}

Should I throw an error or return a rejected promise inside an async function?

They are correct.

The call to myCustomFunction assumes that a promise is returned at all times (.then and .catch deal with resolved and rejected promises, respectively). When you throw an error, the function doesn't return a promise.

You could use this to catch the error:

try {
  myModule.myCustomFunction(someInput).then(result => {
    // carry on
  })
  .catch(err => {
    // do something with the error
  })
} catch(err) {
  ...
}

But as you can see, this results in two error handlers: try/catch for the synchronously thrown error, and .catch for any rejected promises that sns.createTopic(someParams) may return.

That's why it's better to use Promise.reject():

module.exports.myCustomFunction = input => {

    if (badInput) {
        return Promise.reject('failed');
    }

    return sns.createTopic(someParams).promise()
}

Should an Async API Ever Throw Synchronously