Swinject-CodeGen 2.0.0

You want to add pod 'Swinject-CodeGen', '~> 2.0' similar to the following to your Podfile:

target 'MyApp' do
  pod 'Swinject-CodeGen', '~> 2.0'
end

Then run a pod install inside your terminal, or from CocoaPods.app.

Alternatively to give it a test run, run the command:

pod try Swinject-CodeGen

Swinject-CodeGen 2.0.0

By
Swinject Contributors

Swinject Code Generation

Swinject-CodeGen provides a method to get rid of duplicate use of class values and namestrings, by generating explicit functions for registering and resolving using Swinject. Doing this, we also can generate typed tuples to use when resolving, thus allowing better documented and less error-prone code.

Integration

Define your dependencies in a .csv or .yml file (see below and example file)
Add a call to generate the code as build script phase:

For Cocoapods:

$PODS_ROOT/Swinject-CodeGen/bin/swinject_codegen -i baseInput.csv -o extensions/baseContainerExtension.swift

For Carthage:

$SRCROOT/Carthage/Checkouts/Swinject-CodeGen/bin/swinject_codegen -i baseInput.csv -o extensions/baseContainerExtension.swift

Add the generated file (here: extensions/baseContainerExtension.swift) to xcode
Repeat if you need to support multiple targets/have multiple input files.

The code is then generated at every build run.

The Issue

When using Swinject, lots of duplicate definitions appear, whenever we do a

container.register(PersonType.self, name: "initializer") { r in
    InjectablePerson(pet: r.resolve(AnimalType.self)!)
}

let initializerInjection = container.resolve(PersonType.self, name:"initializer")!

the tuple (PersonType.self, name:"initializer") becomes very redundant across the code.

Furthermore, when using arguments, as done in

container.register(AnimalType.self) { _, name in Horse(name: name) }
let horse1 = container.resolve(AnimalType.self, argument: "Spirit") as! Horse

the argument: "Spirit" part is not strictly typed when calling it.

We propose a solution to both these problems by using CodeGeneration

Input Format

Input can be given as .csv or .yml

The call

./swinject_codegen -i example.csv -c

can be used to convert example.csv into example.csv.yml (also works for .yml).

CSV

Basic Structure

Our basic csv structure is defined as follows:

SourceClassName; TargetClassName; Identifier; Argument 1 ... 9

The example above would translate to

PersonType; InjectablePerson; initializer

to generate both a registerPersonType_initializer and a resolvePersonType_initializer function.

See the examples below for more examples.

We decided to use ; as delimiter instead of , to allow the use of tuples as types.

Additional Commands

The ruby parser allows using // and # for comments. Empty lines are ignored and can be used for grouping.

#= <header> can be used to specify additional lines, e.g. #= import KeychainAccess

Dictionaries and Arrays as Parameters

When using typed dictionaries or arrays as parameters, use Array<Type> instead of [Type] and Dictionary<TypeA, TypeB> instead of [TypeA:TypeB]:

PersonType; InjectablePerson; initializer; additionalNames:Array<String>; family:Dictionary<String, String>;

YAML

Example for a .yml definition:

---
HEADERS:
  - import ADependency
DEFINITIONS:
- service: PersonType
  component: InjectablePerson
  name: initializer
- service: PersonType
  component: InjectablePerson
- service: PersonType
  component: PersonType
- service: AnotherPersonType
  component: AnotherPersonType
- service: PersonType
  component: InjectablePerson
  arguments:
  - argument_name: argument_name
    argument_type: argument_type
- service: PersonType
  component: InjectablePerson
  arguments:
  - argument_name: argument_name
    argument_type: argument_type
  - argument_name: argument_typewithoutspecificname
    argument_type: argument_typeWithoutSpecificName
  - argument_name: title
    argument_type: String
  - argument_name: string
    argument_type: String
- service: PersonType
  component: InjectablePerson
  name: initializer
  arguments:
  - argument_name: argument_name
    argument_type: argument_type
  - argument_name: argument_typewithoutspecificname
    argument_type: argument_typeWithoutSpecificName
  - argument_name: title
    argument_type: String
  - argument_name: string
    argument_type: String

Generation Examples

Example A: Same class as source and target

Input

PersonType

Output

// this code is autogenerated, do not modify!

import Swinject

extension Resolver {

    func resolvePersonType() -> PersonType {
        return self.resolve(PersonType.self)!
    }
}

extension Container {

    @discardableResult func registerPersonType(registerClosure: @escaping (_ resolver: Resolver) -> (PersonType)) -> ServiceEntry<PersonType> {
        return self.register(PersonType.self, factory: registerClosure)
    }
}

Example B: Different source and target

Input

PersonType; InjectablePerson

Output

// this code is autogenerated, do not modify!

import Swinject

extension Resolver {

    func resolveInjectablePerson() -> InjectablePerson {
        return self.resolve(PersonType.self) as! InjectablePerson
    }
}

extension Container {

    @discardableResult func registerInjectablePerson(registerClosure: @escaping (_ resolver: Resolver) -> (InjectablePerson)) -> ServiceEntry<PersonType> {
        return self.register(PersonType.self, factory: registerClosure)
    }
}

Example C: Different source and target class with name

Input

PersonType; InjectablePerson; initializer

Output

// this code is autogenerated, do not modify!

import Swinject

extension Resolver {

    func resolveInjectablePerson_initializer() -> InjectablePerson {
        return self.resolve(PersonType.self, name: "initializer") as! InjectablePerson
    }
}

extension Container {

    @discardableResult func registerInjectablePerson_initializer(registerClosure: @escaping (_ resolver: Resolver) -> (InjectablePerson)) -> ServiceEntry<PersonType> {
        return self.register(PersonType.self, name: "initializer", factory: registerClosure)
    }
}

Example D: Different source and target with a single, explicitly named argument

Input

PersonType; InjectablePerson; ; argumentName:ArgumentType

Output

// this code is autogenerated, do not modify!

import Swinject

extension Resolver {

    func resolveInjectablePerson(argumentName: ArgumentType) -> InjectablePerson {
        return self.resolve(PersonType.self, argument: argumentName) as! InjectablePerson
    }
}

extension Container {

    @discardableResult func registerInjectablePerson(registerClosure: @escaping (_ resolver: Resolver, _ argumentName: ArgumentType) -> (InjectablePerson)) -> ServiceEntry<PersonType> {
        return self.register(PersonType.self, factory: registerClosure)
    }
}

Example E: Different source and target with multiple arguments, both explicitly named and not

If no explicit name is given, the lowercase type is used as argumentname.

Input

PersonType; InjectablePerson; ; argumentName:ArgumentType; ArgumentTypeWithoutSpecificName; title:String; String

Output

// this code is autogenerated, do not modify!

import Swinject

extension Resolver {

    func resolveInjectablePerson(argumentName: ArgumentType, argumenttypewithoutspecificname: ArgumentTypeWithoutSpecificName, title: String, string: String) -> InjectablePerson {
        return self.resolve(PersonType.self, arguments: argumentName, argumenttypewithoutspecificname, title, string) as! InjectablePerson
    }
}

extension Container {

    @discardableResult func registerInjectablePerson(registerClosure: @escaping (_ resolver: Resolver, _ argumentName: ArgumentType, _ argumenttypewithoutspecificname: ArgumentTypeWithoutSpecificName, _ title: String, _ string: String) -> (InjectablePerson)) -> ServiceEntry<PersonType> {
        return self.register(PersonType.self, factory: registerClosure)
    }
}

Example F: Different source and target with name with multiple arguments, both explicitly named and not

Input

PersonType; InjectablePerson; initializer; argumentName:ArgumentType; ArgumentTypeWithoutSpecificName; title:String; String

Output

// this code is autogenerated, do not modify!

import Swinject

extension Resolver {

    func resolveInjectablePerson_initializer(argumentName: ArgumentType, argumenttypewithoutspecificname: ArgumentTypeWithoutSpecificName, title: String, string: String) -> InjectablePerson {
        return self.resolve(PersonType.self, name: "initializer", arguments: argumentName, argumenttypewithoutspecificname, title, string) as! InjectablePerson
    }
}

extension Container {

    @discardableResult func registerInjectablePerson_initializer(registerClosure: @escaping (_ resolver: Resolver, _ argumentName: ArgumentType, _ argumenttypewithoutspecificname: ArgumentTypeWithoutSpecificName, _ title: String, _ string: String) -> (InjectablePerson)) -> ServiceEntry<PersonType> {
        return self.register(PersonType.self, name: "initializer", factory: registerClosure)
    }
}

Usage Examples

Using the examples given at the beginning, we can now instead of

container.register(PersonType.self, name: "initializer") { r in
    InjectablePerson(pet: r.resolve(AnimalType.self)!)
}

let initializerInjection = container.resolve(PersonType.self, name:"initializer")!

write:

container.registerPersonType_initializer { r in
    InjectablePerson(pet: r.resolve(AnimalType.self)!)
}

let initializerInjection = container.resolvePersonType_initializer()

Also

container.register(AnimalType.self) { _, name in Horse(name: name) }
let horse1 = container.resolve(AnimalType.self, argument: "Spirit")

becomes

container.registerAnimalType { (_, name:String) in
  Horse(name: name)
}
let horse1 = container.resolveAnimalType("Spirit")

Migration

The script also generates migration.sh files (when using the -m switch), which use sed to go through the code and replace simple cases (i.e. no arguments) of resolve and register. No automatic migration is available for cases with arguments, yet. Simply call the .sh file from the root of the project and compare the results in a git-GUI.

Results

We currently use the code generation in two medium-sized apps across tvOS and iOS.

We found our code to become much more convenient to read and write, due to reduced duplication and autocompletion. We also have a much better overview the classes available through dependency injection. Changing some definition immediately leads to information, where an error will occur. We were able to replace all our occurences of .resolve( and .register( using the current implementation.

Contributors

The original idea for combining CodeGeneration and Swinject came from Daniel Dengler, David Kraus and Wolfgang Lutz.