Entry point can be marked with the 'async' modifier on CoreCLR?

问题

In Stephan Cleary's recent blog post about Async Console Apps on .NET CoreCLR he shows us that in CoreCLR (currently running on Visual Studio 2015, CTP6), the entry point "Main" can actually be marked as async Task, compile properly and actually run:

public class Program
{
    public async Task Main(string[] args)
    {
        Console.WriteLine("Hello World");
        await Task.Delay(TimeSpan.FromSeconds(1));
        Console.WriteLine("Still here!");
        Console.ReadLine();
    }
}

Gives the following output:

This is strengthend by a blog post from the ASP.NET team called A Deep Dive into the ASP.NET 5 Runtime:

In addition to a static Program.Main entry point, the KRE supports instance-based entry points. You can even make the main entry point asynchronous and return a Task. By having the main entry point be an instance method, you can have services injected into your application by the runtime environment.

We know that up until now, An entry point cannot be marked with the 'async' modifier. So, how is that actually possible in the new CoreCLR runtime?

回答1:

Diving into the source of the CoreCLR runtime, we can see a static class called RuntimeBootstrapper, which is in charge of invoking our entry point:

public static int Execute(string[] args)
{
    // If we're a console host then print exceptions to stderr
    var printExceptionsToStdError = Environment.GetEnvironmentVariable(EnvironmentNames.ConsoleHost) == "1";

    try
    {
        return ExecuteAsync(args).GetAwaiter().GetResult();
    }
    catch (Exception ex)
    {
        if (printExceptionsToStdError)
        {
            PrintErrors(ex);
            return 1;
        }

        throw;
    }
}

We can see that internally, it calls ExecuteAsync(args).GetAwaiter().GetResult();, which is semantically equivallent to calling Task.Result, except that instead of recieveing the wrapped AggregationException, we recieve the exception unwrapped.

This is important to understand, as there is no "black magic" as to how it's happening. For the current version of the CoreCLR runtime, the method is allowed to marked async Task because it's blocked higher up the callchain by the runtime.

Side Notes:

Diving into ExecuteAsync, we'll see that it ends up calling:

return bootstrapper.RunAsync(app.RemainingArguments);

When looking inside, we see that actual MethodInfo invocation of our entry point:

public static Task<int> Execute(Assembly assembly, string[] args, IServiceProvider serviceProvider)
{
    object instance;
    MethodInfo entryPoint;

    if (!TryGetEntryPoint(assembly, serviceProvider, out instance, out entryPoint))
    {
        return Task.FromResult(-1);
    }

    object result = null;
    var parameters = entryPoint.GetParameters();

    if (parameters.Length == 0)
    {
        result = entryPoint.Invoke(instance, null);
    }
    else if (parameters.Length == 1)
    {
        result = entryPoint.Invoke(instance, new object[] { args });
    }

    if (result is int)
    {
        return Task.FromResult((int)result);
    }

    if (result is Task<int>)
    {
        return (Task<int>)result;
    }

    if (result is Task)
    {
        return ((Task)result).ContinueWith(t =>
        {
            return 0;
        });
    }

    return Task.FromResult(0);
}

来源：https://stackoverflow.com/questions/28938582/entry-point-can-be-marked-with-the-async-modifier-on-coreclr

标签

.net

async-await

coreclr