Inject provides dependency injection for Go. For small Go applications, manually constructing all required objects is more than sufficient. But for large, modular code bases, dependency injection can alleviate a lot of boilerplate.
In particular, Inject provides a simple way of wiring together modular applications. Each module contains configuration and logic to create the objects it provides. The main application installs all of these modules, then calls its main entry point using the injector. The injector resolves any dependencies of the main function and injects them.
- Example usage
- Value bindings
- Singletons
- Literals
- Mapping bindings
- Sequence bindings
- Named bindings
- Interfaces
- Modules
- Validation
The following example illustrates a simple modular application.
First, the main package installs configured modules and calls an entry point:
package main
func run(db *mgo.Database, log *log.Logger) {
log.Println("starting application")
// ...
}
func main() {
injector := New()
injector.Install(
&MongoModule{URI: "mongodb:https://db1.example.net,db2.example.net:2500/?replicaSet=test&connectTimeoutMS=300000"""},
&LoggingModule{Flags: log.Ldate | log.Ltime | log.Llongfile},
)
injector.Call(run)
}Next we have a simple Mongo module with a configurable URI:
package db
type MongoModule struct {
URI string
}
func (m *MongoModule) ProvideMongoDB() (*mgo.Database, error) {
return mgo.Dial(m.URI)
}The logging package shows idiomatic use of inject; it is just a thin wrapper around normal Go constructors. This is the least invasive way of using injection, and preferred.
package logging
// LoggingModule provides a *log.Logger that writes log lines to a Mongo collection.
type LoggingModule struct {
Flags int
}
func (l *LoggingModule) ProvideMongoLogger(db *mgo.Database) *log.Logger {
return NewMongoLogger(db, l.Flags)
}
type logEntry struct {
Text string `bson:"text"`
}
func NewMongoLogger(db *mgo.Database, flags int) *log.Logger {
return log.New(&mongologWriter{c: db.C("logs")}, "", flags)
}
type mongoLogWriter struct {
buf string
c *mgo.Collection
}
func (m *mongoLogWriter) Write(b []byte) (int, error) {
m.buf = m.buf + string(b)
for {
eol := strings.Index(m.buf, "\n")
if eol == -1 {
return len(b), nil
}
line := m.buf[:eol]
err := m.c.Insert(&logEntry{line})
if err != nil {
return len(b), err
}
m.buf = m.buf[eol:]
}
}The simplest form of binding simply binds a value directly:
injector.Bind(http.DefaultServeMux)Function bindings are not singleton by default. For example, the following function binding will be called each time an int is requested:
value := 0
injector.Bind(func() int {
value++
return value
})Wrap the function in Singleton() to ensure it is called only once:
injector.Bind(Singleton(func() int {
return 10
}))To bind a function as a value, use Literal:
injector.Bind(Literal(fmt.Sprintf))Mappings can be bound explicitly:
injector.Bind(Mapping(map[string]int{"one": 1}))
injector.Bind(Mapping(map[string]int{"two": 2}))
injector.Bind(Mapping(func() map[string]int { return map[string]int{"three": 3} }))
injector.Call(func(m map[string]int) {
// m == map[string]int{"one": 1, "two": 2, "three": 3}
})Or provided via a Provider method that includes the term Mapping in its name:
func (m *MyModule) ProvideStringIntMapping() map[string]int {
return map[string]int{"one": 1, "two": 2}
}Sequences can be bound explicitly:
injector.Bind(Sequence([]int{1, 2}))
injector.Bind(Sequence([]int{3, 4}))
injector.Bind(Sequence(func() []int { return []int{5, 6} }))
injector.Call(func(s []int) {
// s = []int{1, 2, 3, 4, 5, 6}
})Or provided via a Provider method that includes the term Sequence in its name:
func (m *MyModule) ProvideIntSequence() []int {
return []int{1, 2}
}The equivalent of "named" values can be achieved with type aliases:
type UserName string
injector.Bind(UserName("Bob"))
injector.Call(func (username UserName) {})Interfaces can be explicitly bound to implementations:
type stringer string
func (s stringer) String() string { return string(s) }
injector.BindTo((*fmt.Stringer)(nil), stringer("hello"))
injector.Call(func(s fmt.Stringer) {
fmt.Println(s.String())
})However, if an explicit interface binding is not present, any bound object implementing that interface will be used:
injector.Bind(stringer("hello"))
injector.Call(func(s fmt.Stringer) {
fmt.Println(s.String())
})Similarly, if sequences/maps of interfaces are injected, explicit bindings will be used first, then inject will fallback to sequences/maps of objects implementing that interface.
Similar to injection frameworks in other languages, inject includes the concept of modules. A module is a struct whose methods are providers. This is useful for grouping configuration data together with providers.
Any method starting with "Provide" will be bound as a Provider. If the method name contains "Multi" it will not be a singleton provider. If the method name contains "Sequence" it will contribute to a sequence of its return type. Similarly, if the method name contains "Mapping" it will contribute to a mapping of its return type.
type MyModule struct {}
// Singleton provider.
func (m *MyModule) ProvideLog() *log.Logger { return log.New() }
type Randomness int
// "Multi" provider, called every time an "int" is injected.
func (m *MyModule) ProvideMultiRandomness() Randomness { return Randomness(rand.Int()) }Finally, after binding all of your types to the injector you can validate that
a function is constructible via the Injector by calling Validate(f).
Or you can live on the edge and simply use Call(f) which will panic if
injection is not possible.