Github
文档
Dependency injection framework for go programs (golang).
DI handles the life cycle of the objects in your application. It creates them when they are needed, resolves their dependencies, and closes them properly when they are no longer used.
If you do not know if DI could help improve your application, learn more about dependency injection and dependency injection containers:
There is also an Examples section at the end of the documentation.
DI is focused on performance.
Table of contents
Basic usage
A definition contains at least a Build function to create the object.
MyObjectDef := &di.Def{
Build: func(ctn di.Container) (interface{}, error) {
return &MyObject{}, nil
},
}
// It is possible to add a name or a type to make the definition easier to retrieve.
// But it is not mandatory. Check the "Definitions" part of the documentation to learn more about that.
MyObjectDef.SetName("my-object")
MyObjectDef.SetIs((*MyObject)(nil))
The definition can be added to a builder with the Add method:
builder, err := di.NewEnhancedBuilder()
err = builder.Add(MyObjectDef)
Once all the definitions are added to the Builder, you can call the Build method to generate a Container.
ctn, err := builder.Build()
Objects can then be retrieved from the container:
// Either with the definition (recommended)
ctn.Get(MyObjectDef).(*MyObject)
// Or the name (which is slower)
ctn.Get("my-object").(*MyObject)
// Or the type (even slower)
ctn.Get(reflect.typeOf((*MyObject)(nil))).(*MyObject)
The Get method returns an interface{}. You need to cast the interface before using the object.
The container will only call the definition Build function the first time the Get method is called. After that, the same object is returned (unless the definition has its Unshared field set to true). That means the three calls in the example above return the same pointer. Check the Definitions section to learn more about them.
Builder
EnhancedBuilder usage
You need a builder to create a container.
You should use the EnhancedBuilder. It was introduced to add features that could not be added to the original Builder without breaking backward compatibility.
You need to use the NewEnhancedBuilder function to create the builder. Then you register the definitions with the Add method.
If you add two definitions with the same name, the first one is replaced.
builder, err := di.NewEnhancedBuilder()
// Adding a definition named "my-object".
err = builder.Add(&di.Def{
Name: "my-object",
Build: func(ctn di.Container) (interface{}, error) {
return &MyObject{Value: "A"}, nil
},
})
// Replacing the definition named "my-object".
err = builder.Add(&di.Def{
Name: "my-object",
Build: func(ctn di.Container) (interface{}, error) {
return &MyObject{Value: "B"}, nil
},
})
ctn, err := builder.Build()
ctn.Get("my-object").(*MyObject).Value // B
Be sure to handle the errors properly even if it is not the case in this example for conciseness.
EnhancedBuilder limitations
It is only possible to call the EnhancedBuilder.Build function once. After that, it will return an error.
Also, it is not possible to use the same definition in two different EnhancedBuilder.
And you should not update a definition once it has been added to the builder.
All these restrictions exist because the EnhancedBuilder.Build function alters the definitions. It resets the definition fields to their value at the time when the definition was added to the builder. Thus the definitions are linked to the builder and to the container it generates.
Definitions
Definition Build function
A definition only requires a Build function. It is used to create the object.
// You can either use the structure directly.
&di.Def{
Build: func(ctn di.Container) (interface{}, error) {
return &MyObject{}, nil
},
}
// Or use the NewDef function to create the definition.
di.NewDef(func(ctn di.Container) (interface{}, error) {
return &MyObject{}, nil
})
The Build function returns the object and an error if it can not be created.
panics in Build functions are recovered and work as if an error was returned.
Definition dependencies
The Build function can also use the container. This allows you to build objects that depend on other objects defined in the container.
MyObjectDef := di.NewDef(func(ctn di.Container) (interface{}, error) {
return &MyObject{}, nil
})
MyObjectWithDependencyDef := di.NewDef(func(ctn di.Container) (interface{}, error) {
// Using the Get method inside the build function is safe.
// Panics in this function are recovered.
// But be sure to add a name to the definitions if you want understandable error messages.
return &MyObjectWithDependency{
Object: ctn.Get(MyObjectDef).(*MyObject),
}, nil
})
You can not create a cycle in the definitions (A needs B and B needs A). If that happens, an error will be returned at the time of the creation of the object.
Definition name
You can add a name to the definition. It allows you to retrieve the definition from its name.
// Create a definition with a name.
MyObjectDef := &di.Def{
Name: "my-object",
Build: func(ctn di.Container) (interface{}, error) {
return &MyObject{}, nil
},
}
// The SetName method can also be used.
MyObjectDef.SetName("my-object")
// Retrieve the definition from the container.
ctn.Get("my-object").(*MyObject)
If you do not provide a name, a name will be automatically generated when the container is created.
:warning: Names are used in error messages. So it is recommended to set your own names to avoid troubles when debugging.
Retrieving an object from its name instead of its definition requires an additional lookup in a map[string]int. That makes it significantly slower. If performance is critical for you, you should retrieve objects from their definitions.
Another advantage of using the definitions for object retrieval is that it avoids the risk of a typo in the name.
The drawback is that you need to import the package containing the definitions which may lead to import cycles depending on your project structure.
Definition for an already built object
There is a shortcut to create a definition for an object that is already built.
MyObjectDef = di.NewDefFor(myObject)
// is the same as
MyObjectDef = &di.Def{
Build: func(ctn di.Container) (interface{}, error) {
return myObject, nil
},
}
Unshared definitions
By default, the Get method called on the same container always returns the same object.
The object is created when the Get method is called for the first time.
It is then stored inside the container and the same instance is returned in the next calls.
That means that the Build function is only called once.
If you want to retrieve a new instance of the object each time the Get method is called, you need to set the Unshared field of the definition to true.
MyObjectDef = &di.Def{
Unshared: true, // The Build function will be called each time.
Build: func(ctn di.Container) (interface{}, error) {
return &MyObject{}, nil
},
}
// ...
// o1 != o2 because of Unshared=true
o1 := ctn.Get(MyObjectDef).(*MyObject)
o2 := ctn.Get(MyObjectDef).(*MyObject)
Definition Close function
A definition can also have a Close function.
di.Def{
Build: func(ctn di.Container) (interface{}, error) {
return &MyObject{}, nil
},
Close: func(obj interface{}) error {
// Assuming that MyObject has a Close method that returns an error on failure.
return obj.(*MyObject).Close()
},
}
This function is called when the container is deleted.
The deletion of the container must be triggered manually by calling the Delete method.
// Create the Container.
app, err := builder.Build()
// Retrieve an object.
obj := app.Get("my-object").(*MyObject)
// Delete the Container, the Close function will be called on obj.
err = app.Delete()
Definition types
It is possible to set the type of the object generated by the Build function.
It is only declarative and no checks are done to ensure that this information is valid.
It can be used to retrieve an object by its type instead of its name.
You can set multiple types, for example, a structure and an interface implemented by this structure.
MyObjectDef = di.NewDefFor(myObject)
// Declare that myObject is an instance of *MyObject and implements MyInterface.
MyObjectDef.SetIs((*MyObject)(nil), (MyInterface)(nil))
// ...
// Retrieve the object from the types.
ctn.Get(reflect.TypeOf((*MyObject)(nil))).(*MyObject)
ctn.Get(reflect.TypeOf((MyInterface)(nil))).(MyInterface)
:warning: If multiple definitions have the same type, the one that was added last in the builder is used to retrieve the object.
It is possible to use the NewBuildFuncForType function to generate a Build function for a given structure (or pointer to a structure). When the object is created using reflection, it will try to set the fields based on their types and the other definitions. There is also a shortcut NewDefForType to create a definition based on NewBuildFuncForType.
// Definition for an already built object, declared having the type *MyObject.
MyObjectDef = di.NewDefFor(myObject).SetIs((*MyObject)(nil))
// The definition can create a *MyObjectWithDependency
// and the MyObjectWithDependency.Object field will be filled with an object
// from the container if there is one with the same type.
// NewDefForType does not set the type of the definition. You need to call SetIs yourself if you want to.
MyObjectWithDependencyDef := di.NewDefForType((*MyObjectWithDependency)(nil))
// ...
// o1 == o2
o1 := ctn.Get(MyObjectWithDependencyDef).(*MyObjectWithDependency).Object
o2 := ctn.Get(MyObjectDef).(*MyObject)
:warning: It is not recommended to use this because it is hard to know which fields are set and how. In addition to that, the use of reflection in the generated Build function makes it very slow. The behavior of the NewBuildFuncForType may also change in the future if ways to improve the feature are found.
Definition tags
You can add tags to a definition. Tags are not used internally by this library. They are only there to help you organize your definitions.
MyObjectDef = di.NewDefFor(myObject)
tag := di.Tag{
Name: "my-tag",
Args: map[string]string{
"tag-argument": "argument-value",
},
Data: "Data is an interface{} if Args are not enough",
}
MyObjectDef.SetTags(tag)
MyObjectDef.Tags[0] == tag // true
Object retrieval
When a container is asked to retrieve an object, it starts by checking if the object has already been created. If it has, the container returns the already-built instance of the object. Otherwise, it uses the Build function of the associated definition to create the object. It returns the object, but also keeps a reference to be able to return the same instance if the object is requested again (unless the definition is UnShared).
A container can only build objects defined in the same scope (scopes documentation). If the container is asked to retrieve an object that belongs to a different scope. It forwards the request to its parent.
There are three methods to retrieve an object: Get, SafeGet and Fill.
Get
Get returns an interface that can be cast afterward. If the object can not be created, the Get function panics.
// Retrieve the object from the definition (recommended)
o1 := ctn.Get(MyObjectDef).(*MyObject)
// Or from its name (which is slower)
o2 := ctn.Get("my-object").(*MyObject)
// Or from its type (even slower)
o3 := ctn.Get(reflect.typeOf((*MyObject)(nil))).(*MyObject)
// o1 == o2 == o3
SafeGet
Get is an easy way to retrieve an object. The problem is that it can panic. If it is a problem for you, you can use SafeGet. Instead of panicking, it returns an error.
objectInterface, err := ctn.SafeGet(MyObjectDef)
// You still need to cast the interface.
object, ok := objectInterface.(*MyObject)
// SafeGet can also be called with a definition name or type.
objectInterface, err = ctn.SafeGet("my-object")
objectInterface, err = ctn.SafeGet(reflect.typeOf((*MyObject)(nil)))
Fill
The third method to retrieve an object is Fill. It returns an error if something goes wrong like SafeGet, but it may be more practical in some situations. It uses reflection to fill the given object. Using reflection makes it slower than SafeGet.
var object *MyObject
err := ctn.Fill(MyObjectDef, &object)
// Fill can also be called with a definition name or type.
err = ctn.Fill("my-object", &object)
err = ctn.Fill(reflect.typeOf((*MyObject)(nil)), &object)
Scopes
The principle
Definitions can also have a scope. They can be useful in request-based applications, such as a web application.
MyObjectDef := &di.Def{
Scope: di.Request,
Build: func(ctn di.Container) (interface{}, error) {
return &MyObject{}, nil
},
}
The available scopes are defined when the Builder is created:
builder, err := di.NewEnhancedBuilder(di.App, di.Request)
Scopes are defined from the most generic to the most specific (eg: App ≻ Request ≻ SubRequest). If no scope is given to NewEnhancedBuilder, the builder is created with the three default scopes: di.App, di.Request and di.SubRequest. These scopes should be enough almost all the time.
The containers belong to one of these scopes. A container may have a parent in a more generic scope and children in a more specific scope. The builder generates a container in the most generic scope. Then the container can generate children in the next scope thanks to the SubContainer method.
A container is only able to build objects defined in its own scope, but it can retrieve objects in a more generic scope thanks to its parent. For example, a Request container can retrieve an App object, but an App container can not retrieve a Request object.
If a definition does not have a scope, the most generic scope will be used.
Scopes in practice
// Create a Builder with the default scopes (App, Request, SubRequest).
builder, err := di.NewEnhancedBuilder()
// Define an object in the App scope.
AppDef := di.Def{
Scope: di.App, // this line is optional, di.App is the default scope
Build: func(ctn di.Container) (interface{}, error) {
return &MyObject{}, nil
},
}
err =
