ProcedureKit
A Swift framework inspired by WWDC 2015 Advanced NSOperations session. Previously known as Operations, developed by @danthorpe with a lot of help from our fantastic community.
Resource | Where to find it |
---|---|
Session video | developer.apple.com |
Old but more complete reference documentation | docs.danthorpe.me/operations |
Updated but not yet complete reference docs | procedure.kit.run/development |
Programming guide | operations.readme.io |
Compatibility
ProcedureKit supports all current Apple platforms. The minimum requirements are:
- iOS 9.0+
- macOS 10.11+
- watchOS 3.0+
- tvOS 9.2+
The current released version of ProcedureKit (5.1.0) supports Swift 4.2+ and Xcode 10.1. The development
branch is Swift 5 and Xcode 10.2 compatible.
Framework structure
ProcedureKit is a "multi-module" framework (don't bother Googling that, I just made it up). What I mean, is that the Xcode project has multiple targets/products each of which produces a Swift module. Some of these modules are cross-platform, others are dedicated, e.g. ProcedureKitNetwork
vs ProcedureKitMobile
.
Installing ProcedureKit
See the Installing ProcedureKit guide.
Usage
Procedure
is a Foundation.Operation
subclass. It is an abstract class which must be subclassed.
import ProcedureKit
class MyFirstProcedure: Procedure {
override func execute() {
print("Hello World")
finish()
}
}
let queue = ProcedureQueue()
let myProcedure = MyFirstProcedure()
queue.add(procedure: myProcedure)
the key points here are:
- Subclass
Procedure
- Override
execute
but do not callsuper.execute()
- Always call
finish()
after the work is done, or if the procedure is cancelled. This could be done asynchronously. - Add procedures to instances of
ProcedureQueue
.
Observers
Observers are attached to a Procedure
subclass. They receive callbacks when lifecycle events occur. The lifecycle events are: did attach, will execute, did execute, did cancel, will add new operation, did add new operation, will finish and did finish.
These methods are defined by a protocol, so custom classes can be written to conform to multiple events. However, block based methods exist to add observers more naturally. For example, to observe when a procedure finishes:
myProcedure.addDidFinishBlockObserver { procedure, errors in
procedure.log.info(message: "Yay! Finished!")
}
The framework also provides BackgroundObserver
, TimeoutObserver
and NetworkObserver
.
See the wiki on [[Observers|Observers]] for more information.
Conditions
Conditions are attached to a Procedure
subclass. Before a procedure is ready to execute it will asynchronously evaluate all of its conditions. If any condition fails, it finishes with an error instead of executing. For example:
myProcedure.add(condition: BlockCondition {
// procedure will execute if true
// procedure will be ignored if false
// procedure will fail if error is thrown
return trueOrFalse // or throw AnError()
}
Conditions can be mutually exclusive. This is akin to a lock being held preventing other operations with the same exclusion being executed.
The framework provides the following conditions: AuthorizedFor
, BlockCondition
, MutuallyExclusive
, NegatedCondition
, NoFailedDependenciesCondition
, SilentCondition
and UserConfirmationCondition
(in ProcedureKitMobile).
See the wiki on [[Conditions|Conditions]], or the old programming guide on Conditions| for more information.
Capabilities
A capability represents the application’s ability to access device or user account abilities, or potentially any kind of gated resource. For example, location services, cloud kit containers, calendars etc or a webservice. The CapabiltiyProtocol
provides a unified model to:
- Check the current authorization status, using
GetAuthorizationStatusProcedure
, - Explicitly request access, using
AuthorizeCapabilityProcedure
- Both of the above as a condition called
AuthorizedFor
.
For example:
import ProcedureKit
import ProcedureKitLocation
class DoSomethingWithLocation: Procedure {
override init() {
super.init()
name = "Location Operation"
add(condition: AuthorizedFor(Capability.Location(.whenInUse)))
}
override func execute() {
// do something with Location Services here
finish()
}
}
ProcedureKit provides the following capabilities: Capability.CloudKit
and Capability.Location
.
In Operations, (a previous version of this framework), more functionality existed (calendar, health, photos, address book, etc), and we are still considering how to offer these in ProcedureKit.
See the wiki on [[Capabilities|Capabilities]], or the old programming guide on Capabilities for more information.
Logging
Procedure
has its own internal logging functionality exposed via a log
property:
class LogExample: Procedure {
override func execute() {
log.info("Hello World!")
finish()
}
}
See the programming guide for more information on logging and supporting 3rd party log frameworks.
Dependency Injection
Often, procedures will need dependencies in order to execute. As is typical with asynchronous/event based applications, these dependencies might not be known at creation time. Instead they must be injected after the procedure is initialised, but before it is executed. ProcedureKit supports this via a set of protocols and types which work together. We think this pattern is great, as it encourages the composition of small single purpose procedures. These can be easier to test and potentially enable greater re-use. You will find dependency injection used and encouraged throughout this framework.
Anyway, firstly, a value may be ready or pending. For example, when a procedure is initialised, it might not have all its dependencies, so they are in a pending state. Hopefully they become ready by the time it executes.
Secondly, if a procedure is acquiring the dependency required by another procedure, it may succeed, or it may fail with an error. Therefore there is a simple Result type which supports this.
Thirdly, there are protocols to define the input
and output
properties.
InputProcedure
associates an Input
type. A Procedure
subclass can conform to this to allow dependency injection. Note, that only one input
property is supported, therefore, create intermediate struct types to contain multiple dependencies. Of course, the input
property is a pending value type.
OutputProcedure
exposes the Output
associated type via its output
property, which is a pending result type.
Bringing it all together is a set of APIs on InputProcedure
which allows chaining dependencies together. Like this:
import ProcedureKitLocation
// This class is part of the framework, it
// conforms to OutputProcedure
let getLocation = UserLocationProcedure()
// Lets assume we've written this, it
// conforms to InputProcedure
let processLocation = ProcessUserLocation()
// This line sets up dependency & injection
// it automatically handles errors and cancellation
processLocation.injectResult(from: getLocation)
// Still need to add both procedures to the queue
queue.add(procedures: getLocation, processLocation)
In the above, it is assumed that the Input
type matched the Output
type, in this case, CLLocation
. However, it is also possible to use a closure to massage the output type to the required input type, for example:
import ProcedureKitLocation
// This class is part of the framework, it
// conforms to OutputProcedure
let getLocation = UserLocationProcedure()
// Lets assume we've written this, it
// conforms to InputProcedure, and
// requires a CLLocationSpeed value
let processSpeed = ProcessUserSpeed()
// This line sets up dependency & injection
// it automatically handles errors and cancellation
// and the closure extracts the speed value
processLocation.injectResult(from: getLocation) { $0.speed }
// Still need to add both procedures to the queue
queue.add(procedures: getLocation, processLocation)
Okay, so what just happened? Well, the injectResult
API has a variant which accepts a trailing closure. The closure receives the output value, and must return the input value (or throw an error). So, { $0.speed }
will return the speed property from the user's CLLocation
instance.
Key thing to note here is that this closure runs synchronously. So, it's best to not put anything onerous onto it. If you need to do more complex data mappings, check out TransformProcedure
and AsyncTransformProcedure
.
See the programming guide on Injecting Results for more information.