magic.signals 10.0.21

A Super Signal implementation for Hyperlambda

magic.signals is a "Super Signals" implementation for .Net 6 built on top of magic.node, allowing you to invoke functions from one assembly in another assembly without having any direct references between your projects.

Rationale

Below you can find a couple of articles written about the idea by yours truly.

• Super Signals - DZone
• Super Scalable Software Systems

The above is made possible by always having a YALOA, allowing us to invoke methods in classes through a "magic string", which references a type in a dictionary, where the string is its key, and the types are dynamically loaded during startup of your AppDomain. Imagine the following code.

[Slot(Name = "foo.bar")]
public class FooBar : ISlot
{
    public void Signal(ISignaler signaler, Node input)
    {
        input.Value = 42;
    }
}

The above declares a "slot" for the signal [foo.bar]. In any other place in our AppDomain we can use an ISignaler instance to Signal the above slot by using something such as the following.

/*
 * This will invoke our Signal method above
 */
var args = new Node();
signaler.Signal("foo.bar", args);

/*
 * The value of args is now 42.
 */
Assert.Equal(42, args.Value);

Notice that there are no shared types between the invoker and the handler, and there are no references necessary to be shared between these two assemblies. This results in an extremely loosely coupled plugin architecture, where you can dynamically add any plugin you wish into your AppDomain, by simply referencing whatever plugin assembly you wish to bring into your AppDomain, and immediately start consuming your plugin's functionality - Or dynamically loading it during runtime for that matter, resulting in that you instantly have access to its slots, without needing to create cross projects references in any ways.

This results in an extremely loosely coupled architecture of related components, where any one component can easily be exchanged with any other component, as long as it is obeying by the implicit interface of the component you're replacing. Completely eliminating "strongly typing", making interchanging components with other components equally simply in a static programming language such as the .Net CLR as providing a function object in JavaScript. In many ways, this results in having all the advantages from a functional programming language in a static programming language, while still keeping the strongly typing around for cases where you need strongly typing - Allowing you to choose which paradigm you want to use, based upon individual use cases, and not having the language and platform dictate your choices in these regards.

The magic.signals implementation uses IServiceProvider to instantiate your above FooBar class when it wants to evaluate your slot. This makes it behave as a good IoC citizen, allowing you to pass in for instance interfaces into your constructor, and have the .Net dependency injection automatically create objects of whatever interface your slot implementation requires.

There is also an async version of the interface, which allows you to declare async slots, transparently letting the runtime choose which implementation to use, depending upon whether or not it is currently in an async execution context or not. Below you can see how to accomplish the same as above, except this time the slot will be invoked within an async context.

[Slot(Name = "foo.bar")]
public class FooBar : ISlotAsync
{
    public Task SignalAsync(ISignaler signaler, Node input)
    {
        input.Value = 42;
        await Task.Yield();
    }
}

It's a common practice to implements slots that recursively invokes other slots, by combining the above two constructs, into one single class. Below is an example.

[Slot(Name = "foo.bar")]
public class FooBar : ISlot, ISlotAsync
{
    // Sync version.
    public void Signal(ISignaler signaler, Node input)
    {
        input.Value = 42;
    }

    // Async version.
    public Task SignalAsync(ISignaler signaler, Node input)
    {
        input.Value = 42;
        await Task.Yield();
    }
}

The above simple example is probably not that useful to implement as an async slot, but for other parts of your code, where you for instance are accessing sockets, HTTP connections, or the file system for that matter - Creating async slots will have huge advantages for your application's ability to scale, and handle multiple simultaneous users and connections. The runtime will "automagically" choose the correct implementation, being synchronous or asynchronous, depending upon which execution context the execution object currently is within.

If your slots recursively invokes other slots, by for instance invoking signaler.Signal("eval", args), you should also implement the async interface, to allow for children lambda objects to be within an async context. This has huge advantages for your application's throughput.

Passing arguments to your slots

The Node class provides a graph object for you, allowing you to automagically pass in any arguments you wish. Notice, the whole idea is to de-couple your assemblies, hence you shouldn't really pass in anything but "native types", such as for instance System.String, System.DateTime, integers, etc. However, most complex POD structures, can also easily be represented using this Node class. The Node class is basically a name/value/children graph object, where the value can be any object, the name a string, and children is a list of children Nodes. In such a way, it provides a more C# friendly graph object, kind of resembling JSON, allowing you to internally within your assemblies, pass in a Node object as your parameters from the point of your signal, to the slot where you handle the signal. The Node POCO class again, is a bi-directional POD instance, allowing you to both pass arguments into the slot, in addition to having the slot return values back to the caller.

If you invoke Signal or SignalAsync from C#, you can optionally pass in a function object that will be executed after the signal has been executed. This is useful for cases where you're creating an async signal invocation, but not invoking it immediately, and rather returning it as a Task to some other parts of your system, to ensure something occurs after the signal has been executed. Below is an example.

var args = new Node();
return signaler.SignalAsync("foo.bar", args, () => { /* ... This will happen AFTER execution of signal ... */ });

Magic Signals a DSL

A lot of the idea behind Magic Signals is that combined with magic.node, and especially its ability to parse Hyperlambda, it becomes a very good foundation for a DSL, or a Domain Specific programming Language implementation, allowing you to easily create your own programming languages, and keywords, based upon Hyperlambda syntax trees. Hyperlambda in this context being the textual representation of your Node hierarchy.

Project website

The source code for this repository can be found at github.com/polterguy/magic.signals, and you can provide feedback, provide bug reports, etc at the same place.

Quality gates