Inversion of Control (IOC) is a general principle of software architecture, that can be applied in different ways. The Wikipedia article on IOC says it is where "the flow of control of a system is inverted in comparison to procedural programming."
That's one of those definitions that's more useful after you already understand IOC, and not as useful when you're trying to figure it out in the first place. It basically says that IOC is in effect in some form in any system where control is inverted from what "normal" is. But what is "normal"?
If you write Javascript in a browser environment, you're already using one form of IOC. Let's look at a simple example to compare "normal" vs. "inverted" control.
Take a simple program, like a shell script, that executes sequentially from beginning to end, maybe reading files, transforming the data, and then outputting the transformed data to stdout or another file. This is the same as the concept of a "main()" in languages like C, C++, Java, etc.
That's a traditional, or "normal", flow of control. Your code is in control and makes decisions (e.g. via conditionals, loops, etc.) about what code to execute when.
When you write Javascript in a browser, your code will typically be structured, at some level, as a set of callback functions attached to browser events. You won't have a "main()", in the traditional sense, but rather, you rely on the browser to invoke your callback functions at the appropriate time. The browser is in control, and makes decisions about when to give control back to your code by invoking your callback functions. The browser may even decide not to call your callbacks at all, depending on the situation.
So, in a typical browser application, the main control is inverted and resides with the browser. Your higher-level application code, is simply a collection of smaller chunks of code, callbacks, and the browser's lower-level main control loop drives the execution of your application.
Now you might be asking yourself "If I'm already doing IOC, why do I need wire.js?". The example above is just one form of IOC. Wire.js provides another important kind of IOC: Dependency Inversion.
Dependency Inversion is a pattern where concrete components do not directly depend on other concrete components, but rather on abstractions and APIs. The concrete dependencies are provided, via a mechanism such as Dependency Injection (more info below) by the environment in which the components are used.
So, the "inversion" in this case, refers to how components' dependencies are satisfied.
This is probably a good time to go read Martin Fowler's well known article on the subject for a more extensive explanation and examples. Jim Weirich's presentation from OSCON 2005 is also an excellent introduction to both Dependency Inversion and Dependency Injection, and well worth reading.
Side note: The term IOC Container is usually a bit of a misnomer. Most IOC Containers focus primarily on providing Dependency Inversion, and so a better name might be "Dependency Inversion Container". Fowler mentions this as well.
Implementing application logic inside components, and composing those components together into a running application are very different activities. Many times, however, they are done at the same time, in the same code. That leads to tightly coupled components that can be impossible to unit test and refactor.
Separating the process of implementing component logic from application composition has some significant benefits:
- It decouples components from each other, making them easier to test and refactor.
- It decouples components from the connection mechanisms, making it easier to change the type of connection (method call, pubsub, advice-based, sync vs. async, etc.) between them.
- It means that the connection mechanism need not be involved in unit testing the component’s logic.
- Having a designated spot for composition gives developers a place to look to understand the overall structure of an application
wire.js is cujo.js’s application composition layer. It provides a well-defined place for creating, configuring, and non-invasively connecting together the components of an application, or a chunk of an application.
Components can be implemented and tested without embedding connection logic and infrastructure. The composition and application logic can be refactored independently, many times without affecting each other at all.
Similar to declarative IOC containers for other platforms, such as Spring Framework for Java, wire has an extensible DSL. The core DSL has a very simple form, and only a handful of top-level keywords. The DSL can be extended by plugins.
Wire specs are declarative, and thus order does not matter. Wire will process the spec in dependency order, regardless of the order in which you declare components. A typical best practice is to order your spec in a logical way, placing related components near each other, and optimizing for readability and understandability.
A wire spec is a Javascript object literal or JSON object that describes a set of components. Wire parses a spec as input and produces the fully realized set of components as output.
In their simplest form, components can be any simple Javascript type, such as Strings, Numbers, Arrays, etc., including RegExp and Date when the spec is a Javascript object literal. More interestingly, components can be things such as AMD modules, constructors, or factory functions that wire.js can use to create instances of larger, more complex components of your application.
Here is the wire spec from the Hello Wire example. You can read a complete walkthrough of this wire spec at the Hello Wire github repo
define({
message: 'I haz been wired',
// Create an instance of the hello-wired module.
helloWired: {
create: {
module: 'app/HelloWire',
args: { $ref: 'dom:first!.hello' }
},
ready: {
sayHello: { $ref: 'message' }
}
},
plugins: [
{ module: 'wire/debug', trace: true },
{ module: 'wire/dom', $ns: 'dom' }
]
});This simple wire spec has three top-level components:
message- a StringhelloWired- an AMD module with module idapp/HelloWired. In this case the module is a constructor function, which wire.js will use to create an object instance.plugins- an Array containing one AMD module to load. This module happens to be a wire.js plugin for referencing DOM nodes--read more on referencing below and in the References section.
The wire spec also contains two references using JSON Referencing-like syntax. The first references a DOM Node by id:
{ $ref: 'dom!hello' }The second references the message String (the first item in the wire spec):
{ $ref: 'message' }When you feed a spec to wire.js, it will create a context containing fully realized versions of the components in the spec. In the Hello Wire case, the context will contain the message String, an instance of the HelloWired object from the app/HelloWire AMD module, and an Array with one element--the wire/dom plugin AMD module.
As the result of processing a spec, wire.js produces a Context. The context is a Javascript Object that contains the all fully realized components that were specified in the wiring spec. The context also has methods for wiring child contexts, resolving references, and destroying the context and all the objects, etc. that were created when it was wired.
Let's look again at the simple wiring spec from the Hello Wire example.
define({
message: 'I haz been wired',
// Create an instance of the hello-wired module.
helloWired: {
create: {
module: 'app/HelloWire',
args: { $ref: 'dom:first!.hello' }
},
ready: {
sayHello: { $ref: 'message' }
}
},
plugins: [
{ module: 'wire/debug', trace: true },
{ module: 'wire/dom', $ns: 'dom' }
]
});Using wire.js as an AMD plugin, we can wire the spec:
curl(['wire!hello-wired-spec'], function(context) {
console.log(context);
// Components are just properties of the wired context
console.log(context.helloWired)
});which creates the context, context, that contains fully realized components:
message- a StringhelloWired- an object created from the AMD moduleapp/HelloWired, whose constructor was passed a DOM node by thewire/domplugin's DOM reference resolver, and whoseinit()function has been called and passed themessageString.plugins- an Array containing a single wire.js plugin,wire/dom.
The wired context has properties for the components from the wiring spec.
In wire.js, there is an implicit hierarchy of contexts, and there is a Root Context that is the ultimate ancestor of all the contexts you create. This context hierarchy acts a lot like a Javascript prototype chain, in that a child contexts can see components from its parent context, can override them, and can have new components not in its parent.
Any context can be used to create a child by calling context.wire(childSpec). Here's an example using the hello-wire.js spec as the parent to create a child:
// First, create the hello-wire context, same as above.
curl(['wire!hello-wired-spec'], function(context) {
console.log(context);
// Use the context to wire a child
context.wire({
// Child context
anotherComponent: {
// Create an instance by calling constructor with no args
create: 'my/other/component',
// Call anotherComponent.sayHowdy(message)
// Message refers to the message String in the parent context
init: {
sayHowdy: { $ref: 'message' }
}
}
}).then(function(childContext) {
console.log(childContext);
// The child can see components in its parent, similar to
// a Javascript prototype
console.log(childContext.helloWired);
// But also has its own components
console.log(childContext.anotherComponent);
// The parent *cannot* see components in the child
console.log(context.anotherComponent); // logs undefined
});
});The childContext will have properties for all the components in its parent context: message, helloWired, and plugins, but will also have the additional component anotherComponent.
Wire.js's core DSL is very small, but can be extended by plugins. For example, there is no builtin handling of DOM Nodes or DOM Events. That functionality is provided the bundled DOM plugins.
Including plugins in a wire spec is simple. Wire scans modules for plugins, so you can simply include them in your spec using the module factory. Although it's not necessary, a good convention is to group plugins together in a plugins array:
plugins: [
{ module: 'wire/debug' },
{ module: 'wire/dom' },
{ module: 'wire/dom/render' },
{ module: 'wire/aop' }
]Plugins may have options, which can be included as properties. For example, to turn on the wire/debug plugin's trace option:
plugins: [
{ module: 'wire/debug', trace: true },
{ module: 'wire/dom' },
{ module: 'wire/dom/render' },
{ module: 'wire/aop' }
]By default, all the factories and facets provided by each plugin are available un-namespaced within the current wire spec. For clarity, and to avoid potential naming conflicts between plugins, you can optionally provide a namespace for some or all plugins in your wire specs, using the $ns option.
When namespaced, all of the factories, facets, and reference resolvers provided by the plugin must be prefixed with the namespace.
The Hello Wire example from above assigns the namespace dom to the wire/dom plugin, and thus uses the plugin's first! resolver with the namespace prefix: dom:first!
define({
message: 'I haz been wired',
// Create an instance of the hello-wired module.
helloWired: {
create: {
module: 'app/HelloWire',
// Use the first! resolver with namespace prefix
args: { $ref: 'dom:first!.hello' }
},
ready: {
sayHello: { $ref: 'message' }
}
},
plugins: [
{ module: 'wire/debug', trace: true },
// Assign the namespace `dom` to the wire/dom plugin
{ module: 'wire/dom', $ns: 'dom' }
]
});One of the main things you'll do when assembling any application, whether you're doing it programmatically in pure Javascript, or declaratively using wire, is to create instances of components--from primitive Javascript types, like Numbers and Strings, to more elaborate things like DOM-based views and Javascript controllers.
Wire.js supports a wide variety of components from simple Javascript types, to object literals and Arrays, to AMD modules.
A component can be any native Javascript type: Number, String, Boolean, Date, RegExp (via both new RegExp and literal // syntax), Array, Object literal.
In addition to simple types, wire uses factories to create more interesting components, such as AMD and CommonJS modules, object instances using constructors or Object.create (in an ES5 environment), functions, etc.
Read more about creating components
Proxies are closely related to factories. For each component, wire creates a proxy that allows other plugins to interact with the component in a generic way.
For example, each proxy implements a simple get()/set() API for getting and setting its component's properties. This allows plugins to set properties on objects where simple property assignment is not sufficient. For example, Dojo Dijit widgets require calling their get() and set() methods.
Each component in a wire spec has a well-defined lifecycle that is managed by wire.js. When wire.js processes the spec to create a context, each component will pass through the following lifecycle stages:
- Create
- Configure
- Initialize
- Connect
- Ready
- Destroy
During the Create stage, a factory creates the component instance. Then, during each lifecycle stage (including Create), various facets will be applied to the instance.
When a context is destroyed by calling its destroy() method, the components will go through a final stage. During the Destroy phase, facets can also be applied, although typically these will be specialized facets that help clean up the component and its resources.
Facets allow you to apply additional configuration or behavior to a component after it has been created by a factory. For example, the properties facet sets properties on a component during the Configure lifecycle stage, the ready facet invokes initializer methods on a component during the Ready stage, and the connect facet (provided by the wire/connect plugin) can connect components together allowing them to collaborate.
Wire.js comes with several builtin facets, and plugins can provide additional facets.
References allow you to reference components and other existing resources. Wire.js uses a JSON-referencing-like syntax for references, but allows for extensions to the referencing syntax via plugins, which can provide Reference Resolvers to do more sophisticated things, such as referencing DOM nodes
In their most simple form, references are a Javascript object with a single property $ref. For example, the following reference refers to a component in the current context(or any of its ancestors) whose name is myComponent.
{ $ref: 'myComponent' }When using plugin reference resolvers, the syntax is similar to AMD loader plugin syntax:
{ $ref: 'resolver!reference-identifier' }For example, the wire/dom plugin provides a reference resolver for referencing DOM nodes by id by providing a reference resolver named dom, whose reference identifier is a DOM node id.
{ $ref: 'dom!my-node-id' }Using references in a wire spec is similar to using variables. For example, if you have a component named controller needs a reference to another component named view:
// Create a controller instance
controller: {
create: 'my/Controller',
// Set controller properties
properties: {
// Set the controller's myView property to the view
// instance created below
myView: { $ref: 'view' }
// ... other controller properties
}
// ... other controller configuration
},
// Create a view instance
view: {
create: 'my/View',
// ... other view configuration
}Notice that order doesn't matter. Even though view is referenced before it is declared, the reference will be resolved correctly because wire specs are declarative, and wire.js will handle ordering to make sure everything works out.
Connecting components together so they can collaborate is at the heart of building any application. Connections are the lines in your "box and line" diagrams. The type of connections as well as how they are created can be just as important as the components they connect.
Many times, application components create their own connections to other components, which can lead to an inflexible architecture that is difficult to refactor.
Wire.js allows you to connect components together non-invasively using several types of connectors, such as dependency injection, synthetic Javascript events, and Aspect-Oriented Programming (AOP). For browser-based applications, it also treats the DOM in the same way. It allows you to bind DOM events between DOM nodes and Javascript components without coding it into your components.
Wire.js connections can also transform data that flows through them, making it easier to mix and match components by not forcing you to put data-adapting logic into the components themselves.
Separating connections from application logic makes for simpler testing, easier refactoring, and a more flexible architecture.