Skip to content

Latest commit

 

History

History
536 lines (378 loc) · 20 KB

README.md

File metadata and controls

536 lines (378 loc) · 20 KB

Cadmium

Cadmium is a Core Data framework for Swift that enforces best practices and raises exceptions for common Core Data pitfalls exactly where you make them.

Cadmium was written as a reaction to the complexity of dealing with multiple managed object contexts for standard database-like use cases. It is still important to understand what a managed object context is and how they are used, but for typical CRUD-style usage of Core Data it is a complete nuisance.

With Cadmium, the user never sees a NSManagedObjectContext or derived class. You interact only with argument-less transactions, and object fetch/manipulation tasks. The contexts are managed in the background, which makes Core Data feel more like Realm.

Design Goals

  • Create a minimalist/concise framework API that provides for most Core Data use cases and guides the user towards best practices.
  • Aggressively protect the user from performing common Core Data pitfalls, and raise exceptions immediately on the offending statement rather than waiting for a context save event.

Here's an example of a Cadmium transaction that gives all of your employee objects a raise:

Cd.transact {
    try! Cd.objects(Employee.self).fetch().forEach {
        $0.salary += 10000
    }
}

You might notice a few things:

  • Transaction usage is dead-simple. You do not declare any parameters for use inside the block.
  • You never have to reference the managed object context, we manage it for you.
  • The changes are committed automatically upon completion (you can disable this.)

What Cadmium is Not

Cadmium is not designed to be a 100% complete wrapper around Core Data. Some of the much more advanced Core Data features are hidden behind the Cadmium API. If you are creating an enterprise-level application that requires meticulous manipulation of Core Data stores and contexts to optimize heavy lifting, then Cadmium is not for you.

Cadmium is for you if want a smart wrapper that vastly simplifies most Core Data tasks and warns you immediately when you inadvertently manipulate data in a way you shouldn't.

Installing

You can install Cadmium by adding it to your CocoaPods Podfile:

pod 'Cadmium'

Or you can use a variety of ways to include the Cadmium.framework file from this project into your own.

Swift Version Support

Swift 3.1: Use Cadmium 1.1

Swift 3.0: Use Cadmium 1.0

Swift 2.3: Use Cadmium 0.13.x

Swift 2.2: Use Cadmium 0.12.x

Cocoapods:

pod 'Cadmium', '~> 1.2'  # Swift 4.0
pod 'Cadmium', '~> 1.1'  # Swift 3.1 
pod 'Cadmium', '~> 1.0'  # Swift 3.0
pod 'Cadmium', '~> 0.13' # Swift 2.3
pod 'Cadmium', '~> 0.12' # Swift 2.2

How to Use

Context Architecture

Cadmium uses the same basic context architecture as CoreStore, with a root save context running on a private queue that has one read-only child context on the main queue and any number of writeable child contexts running on background queues.

Cadmium Core Data Architecture

This means that your main thread will never bog down on write transactions, and will only be used to merge changes (in memory) and updating any UI elements dependent on your data.

It also means that you cannot initiate modifications to managed objects on the main thread! All of your write operations must exist inside transactions that occur in background threads. You will need to design your app to support the idea of asynchronous write operations, which is what you should be doing when it comes to database modification.

Managed Object Model

The creation and use of the managed object model is very similar to typical Core Data flow. Create your managed object model as usual, and generate the corresponding NSManagedObject classes. Then, simply change the hierarchy so that your class implementations derive from CdManagedObject instead of NSManagedObject.

CdManagedObject is a child class of NSManagedObject.

Initialization

Set up Cadmium with a single initialization call:

do {
    try Cd.initWithSQLStore(momdInbundleID: nil,
                            momdName:       "MyObjectModel.momd",
                            sqliteFilename: "MyDB.sqlite",
                            options:        nil /* Optional */)
} catch let error {
    print("\(error)")
}

This loads the object model, sets up the persistent store coordinator, and initializes important contexts.

If your object model is in a framework (not your main bundle), you'll have to pass the framework's bundle identifier to the first argument.

The options argument flows through to the options passed in addPersistentStoreWithType: on the NSPersistentStoreCoordinator.

You can pass nil to the sqliteFilename parameter to create an NSInMemoryStoreType database.

Querying

Cadmium offers a chained query mechanism. This can be used to query objects from the main thread (for read-only purposes), or from inside a transaction.

Querying starts with Cd.objects(..) and looks like this:

do {
    for employee in try Cd.objects(Employee.self)
                          .filter("name = %@", someName)
                          .sorted("salary", ascending: true)
                          // See CdFetchRequest for more functions
                          .fetch() {
        /* Do something */
        print("Employee name: \(employee.name)")
    }
} catch let error {
    print("\(error)")
}

You begin by passing the managed object type into the parameter for Cd.objects(..). This constructs a CdFetchRequest for managed objects of that type.

Chain in as many filter/sort/modification calls as you want, and finalize with fetch() or fetchOne(). fetch() returns an array of objects, and fetchOne() returns a single optional object (nil if none were found matching the filter).

Transactions

You can only initiate changes to your data from inside of a transaction. You can initiate a transaction using either:

Cd.transact {
    //...
}
Cd.transactAndWait {
    //...
}

Cd.transact performs the transaction asynchronously (the calling thread continues while the work in the transaction is performed). Cd.transactAndWait performs the transaction synchronously (it will block the calling thread until the transaction is complete.)

To ensure best practices and avoid potential deadlocks, you are not allowed to call Cd.transactAndWait from the main thread (this will raise an exception.)

Implicit Transaction Commit

When a transaction completes, the transaction context automatically commits any changes you made to the data store. For most transactions this means you do not need to call any additional commit/save command.

If you want to turn off the implicit commit for a transaction (e.g. to perform a rollback and ignore any changes made), you can call Cd.cancelImplicitCommit() from inside the transaction. A typical use case would look like:

Cd.transact {

    modifyThings()

    if someErrorOccurred {
        Cd.cancelImplicitCommit()
        return
    }

    moreActions()
}

You can also force a commit mid-transaction by calling Cd.commit(). You may want to do this during long transactions when you want to save changes before possibly returning with a cancelled implicit commit. A use case might look like:

Cd.transact {

    modifyThingsStepOne()
    Cd.commit() //changes in modifyThingsStepOne() cannot be rolled back!

    modifyThingsStepTwo()

    if someErrorOccurred {
        Cd.cancelImplicitCommit()
        return
    }

    moreActions()
}

Forced Serial Transactions

NOTE: Advanced Feature

Core Data, and Cadmium, are asynchronous APIs by nature. You generally initiate fetches and modify data asynchronously from the main thread. This tight coupling with asynchronous behavior may be detrimental if you find that the context modifications you perform often conflict with each other.

For example, take the following transaction that might occur when the user taps a button to visit a place:

Cd.transact {
    if let place = try! Cd.objects(Place.self).filter("id = %@", myID).fetchOne() {
        place.visits += 1
    }
}

What if the user spams the visit button? Because the transactions occur in separate context queues, it's not 100% guaranteed that place.visits will increment serially. There is a remote possibility that a race condition will cause two of these contexts to see place.visits as the same value before incrementing.

To help resolve this problem and ensure that transactions are executed serially, you may pass an optional serial parameter to your transactions:

Cd.transact(serial: true) {
    if let place = try! Cd.objects(Place.self).filter("id = %@", myID).fetchOne() {
        place.visits += 1
    }
}

This guarantees that your transactions will occur serially (even waiting for the finalized context save) before proceeding to the next one -- thus your transactions can be considered atomic.

Note that this atomic behavior is limited to the top-most transaction. Transactions-inside-transactions are not executed on the serial queue to prevent deadlocks.

/* In this odd case, the inside transactAndWait will ignore the serial parameter, since it is already
   inside a transaction.  The prevents deadlocks waiting on the serial queue. */

Cd.transact(serial: true) {
    Cd.transactAndWait(serial: true) {
    }
}

It is annoying to pass the serial parameter every time you want this behavior, especially if you want in most of the time. If you want your transactions to be serial by default, pass true to the serialTX parameter in the Cd.init function:

try Cd.initWithSQLStore(momdInbundleID: "org.fieldman.CadmiumTests",
                        momdName:       "CadmiumTestModel",
                        sqliteFilename: "test.sqlite",
                        serialTX:       true)

When you pass true to the init, you can override this per-transaction by passing false to the serial parameter:

Cd.transact(serial: false) {
    ...
}

Not specifying the serial parameter, or passing nil, will use the default determined during initialization.

Most use cases will be fine setting the default serial usage to true and leaving it at that. You will incur a performance hit with serial transactions in the cases that you attempt to execute lots of concurrent or long-running transactions on unrelated objects (since these will all execute serially instead of in parallel).

As an advanced workaround to this issue, if you have many concurrent long-running transactions on different subsets of your data store, you can pass your own dispatch queue to the on parameter:

Cd.transact(on: mySerialDispatchQueue) {
    ...
}

The transaction block itself will occur in the context's queue -- but this occurs synchronously in the dispatch queue you provide. You can provide different dispatch queues for transactions that affect different objects.

Note that the dispatch queue you provide will target the internal default serial queue, which keeps the save context synchronized no matter which queues you use. It also means that any transactions you run on the default serial queue will block all other queues you may pass in (so avoid running very long transactions on the default serial queue).

Creating and Deleting Objects

Objects can be created and deleted inside transactions.

Cd.transact {
    let newEmployee    = try! Cd.create(Employee.self)
    newEmployee.name   = "Bob"
    newEmployee.salary = 10000
}

Cd.transact {
    Cd.delete(try! Cd.objects(Employee.self).filter("name = %@", "Bob").fetch())
}

You can also delete objects directly from a CdFetchRequest:

Cd.objects(Employee.self).filter("salary > 100000").delete()

If called:

  • Outside a transaction: will delete the objects asynchronously in a background transaction.
  • Inside a transaction: will perform the delete synchronously inside the transaction.

Modifying Objects from Other Contexts

You will often need to modify a managed object from one context inside of another context. The most common use case is when you want to modify objects you've queried from the main thread (which are read-only).

You can use Cd.useInCurrentContext to get a copy of the object that is suitable for modification in the current context:

/* Acquire a read-only employee object somewhere on the main thread */
guard let employee = try! Cd.objects(Employee.self).fetchOne() else {
    return
}

/* Modify it in a transaction */
Cd.transact {
    guard let txEmployee = Cd.useInCurrentContext(employee) else {
        return
    }

    txEmployee.salary += 10000    
}

Note that an object must have been inserted and committed in a transaction before it can be accessed from another context. If a transient object has not been inserted yet, it will not be available with this method.

If you are only using one object from another context, consider Cd.transactWith instead:

/* Acquire a read-only employee object somewhere on the main thread */
guard let employee = try! Cd.objects(Employee.self).fetchOne() else {
    return
}

/* Modify it in a transaction */
Cd.transactWith(employee) { txEmployee in
    if let txEmployee = txEmployee {
        txEmployee.salary += 10000
    }
}

You can also pass an array into Cd.transactWith to get an array of objects in the new context.

Notifying the Main Thread

Because transactions occur on the transaction context's private queue, calls to Cd.commit() are synchronous and only return after the save has propagated to the persistent store.

You can use this fact to notify the main thread that a commit has completed in your transaction:

Cd.transact {

    modifyThings()
    Cd.commit()

    /* only called after the commit saves up to the persistent store */
    DispatchQueue.main.async {
        notifyOthers()
    }    
}

You must also call Cd.commit() if you want to dispatch objects created in a transaction back to the main thread, since calling Cd.commit() will save created objects to the persistent store and give them permanent IDs.

Cd.transact {
    let newItem = try! Cd.create(ExampleItem.self)
    newItem.name = "Test"

    /* Synchronously saves newItem to the persistent store */
    try! Cd.commit()

    Cd.onMainWith(newItem) { mainItem in
        guard let item = mainItem else {
            return
        }

        print("created item in transaction: \(item.name)")        
    }
}

Fetched Results Controller

For typical uses of NSFetchedResultsController, you should use the built-in subclass CdFetchedResultsController. This subclass wraps the normal functionality of NSFetchedResultsController onto the protected main queue context.

You can use the CdFetchedResultsController as you would a NSFetchedResultsController with the following in mind:

  • The objects in the fetch results exist in the main thread read-only context and cannot be modified. Use Cd.useInCurrentContext to modify them in a transaction.
  • You can pass a UITableView into the automateDelegation method to perform the standard insert/delete commands on sections and rows when your fetched results controller has changes. This can help save a few lines in your own view controllers.

Using the Update Handler

Every instance of CdManagedObject has a property called updateHandler that can store a block to be called when it is updated. You may only attach a block to updateHandler on objects belonging to the main thread context. This can be useful in situations where you want to monitor objects without using an NSFetchedResultsController.

An example might look like:

/* ... from the example above, transferring a new item to the main thread: */
DispatchQueue.main.async {
    if let mainItem = Cd.useInCurrentContext(newItem), name = mainItem.name {
        print("created item in transaction: \(name)")
        mainItem.updateHandler = { event in
            print("event occurred on object \(name): \(event)")
        }
    }
}

Be aware that you can only install one updateHandler per instance. If you need a solution that requires dispatching to more listeners, you can use the handler to post a NSNotification, or use another toolkit like ReactiveCocoa (see the file Cadmium+ReactiveCocoa.swift in the Examples directory.)

Aggressively Identifying Coding Pitfalls

Most developers who use Core Data have gone through the same gauntlet of discovering the various pitfalls and complications of creating a multi-threaded Core Data application.

Even seasoned veterans are still susceptible to the occasional 1570: The operation couldn’t be completed or 13300: NSManagedObjectReferentialIntegrityError

Many of the common issues arise because the standard Core Data framework is lenient about allowing code that does the Wrong Thing and only throwing an error on the eventual attempt to save (which may not be proximal to the offending code.)

Cadmium performs aggressive checking on managed object operations to make sure you are coding correctly, and will raise exceptions on the offending lines rather than waiting for a save to occur.

PromiseKit Extension

You can enable the Cadmium PromiseKit extension by adding

pod 'Cadmium/PromiseKit'

To your Podfile. This will enable the following functionality:

Transaction Promises

With the PromiseKit extension, Cd.transact and Cd.transactWith are now given promise-return overrides. In most cases the compiler should be able to deduce these as long as you treat the return of the transaction like a promise:

Cd.transact {
    //...
}.then { _ -> Void in

}

Cd.transactWith(obj) { txObj in
    //...
}.then { _ -> Void in

}

The implementation is such that the transaction changes are committed before the promise is fulfilled, so the transaction promise is fully atomic from the perspective of the promise chain.

Note that the compiler may need you to be somewhat explicit about the promise chain. If you see odd errors, try explicitly defining the promise signatures instead of letting the compiler try to infer them (e.g. how the example above adds _ -> Void in instead of leaving it empty).

Transactions Inside the Chain

If you would like to use a transaction inside the promise chain, some more options are available to you:

firstly {
    somePromise()
}.thenTransact(serial: ..., on: ...) { // Optionally use serial: and on: arguments
    // This block occurs inside a Cadmium transaction
    // Commit is called before the promise is fulfilled.
}.then {

}

// Or use the transactWith variant when the previous
// promise is fulfilled with a CdManagedObject

firstly {
    return employeeVarFromMainThread // <- CdManagedObject
}.thenTransactWith(serial: ..., on: ...) { (employee: Employee) -> Void in
    // This block occurs inside a Cadmium transaction with a
    // version of the argument belonging to the current context.
}.then {

}

There are times when you need to funnel a CdManagedObject from a transaction back to the main thread. For this, use thenOnMainWith:

firstly {
    return employeeVarFromMainThread
}.thenTransactWith(serial: ..., on: ...) { (employee: Employee) -> Employee in
    employee.salary += 10000
    return employee
}.thenOnMainWith { (employee: Employee) -> Void
    // Here, employee is a read-only CdManagedObject from the main thread context.
    print("The salary is \(employee.salary)")
}

It should be noted that promise block for thenTransactWith can receive an optional in the pipe, but does not pass an optional as its block argument. It is considered a promise error if the fulfillment value to thenTransactWith is nil, or if the internal Cd.useInCurrentContext returns a nil value. In these cases the chain will be rejected with CdPromiseError.NotAvailableInCurrentContext(value).