I know I mentioned we would talk about customizing the fetch requests, however, I have been working on some code related to the Active Record Fetching project, which I am renaming to MagicalRecord, that is also just as useful as fetching–threading.
Whenever most cocoa developers mention threading and Core Data in the same sentence, the reaction I see most often is that of mysticism and disbelief. For one, multithreaded programming in general is hard–hard to design correctly, hard to write correctly, and debugging threads is just asking for trouble. Introducing Core Data into that mix can seem like the straw that broke the camel’s back. However, by following a few simple rules and guidelines, and codifying them into a super simple pattern, one that may be familiar to you, we can achieve safer Core Data threading without the common headaches.
First up, I want to thank Matt Long and Marcus Zarra for allowing me to guest post on CIMGF. This post is the first in a short series of topics describing how to I’ve made using Core Data a little simpler without giving up the power features you still need. The full project from which this series is derived is available on github.
Core Data, for both iPhone and Mac, is a very powerful framework for persisting your objects out of memory, and into a more permanent storage medium. With the enormous power of Core Data, it can be easy to slip into the trap of thinking that Core Data is very complex.
**This article is reprinted from the MDN**
The documentation on Core Data for the iPhone has lead to some confusion about how best to use Core Data on a Cocoa Touch device. One particular section seems to be the most confusing, specifically:
> A view controller typically shouldn’t retrieve the context from a global object such as the application delegate. This tends to make the application architecture rigid. Neither should a view controller typically create a context for its own use. This may mean that operations performed using the controller’s context aren’t registered with other contexts, so different view controllers will have different perspectives on the data.
> When you create a view controller, you pass it a context. You pass an existing context, or (in a situation where you want the new controller to manage a discrete set of edits) a new context that you create for it. It’s typically the responsibility of the application delegate to create a context to pass to the first view controller that’s displayed.
The idea behind this section is the issue of rigidity. Ideally, each view controller should be an island on its own. It should not rely on its parent, nor should it rely on the Application Delegate. Once a view controller is pushed onto the screen it should ideally be its own master.
## Why Rigidity is bad
It is fairly common when designing a Cocoa Touch application to “hard code” everything. Take the following navigation controller design:
![Navigation Controller Design](http://www.cimgf.com/wp-content/uploads/2011/01/Image1.png “Standard Navigation Controller Design”)
When this design, it is common to code each view controller and make it “aware” of its parent. In that design, it would be common to see view controller B call methods or call back (to its delegate) view controller A. While there is nothing technically wrong with this design, it is very rigid. It is nearly impossible to either move view controller B to another location in the stack or to reuse view controller B somewhere else. This is the trap that the documentation is trying to help new developers avoid.
## Solution One
Again using a standard/normal navigation controller design, it is expected that the detail flows from left to right. The left most (or root) view controller contains the most vague information and the right most (or deepest) view controller contains the greatest detail.
In this case then the best solution is to use a `UIFetchedResultsController`. This controller can be considered a thin layer between the view controllers and the Core Data bits. The advantage is that the `UIFetchedResultsController` is designed to work with tables. The other advantage is that your least detailed view (the root most likely) can listen as the delegate of the `UIFetchedResultsController` for changes and update itself.
In this design, however, instead of passing around a context, you would hand off just the entity that the child view controller needs to know about. The Core Data Recipes example provided by Apple illustrates this design quite well.
How does this break rigidity? Each view controller, from the root on down, only knows what is passed into it. The root gets the `UIFetchedResultsController` passed into it. The child views only get the items it cares about passed into it. None of them care what their parent view controller is. There is no call back to a parent.
## Solution two
What happens when we don’t have a typical navigation controller design? Perhaps a child view can pop up a modal view that displays different information. Perhaps a child view, for whatever reason needs to access information that cannot be directly passed into it every time.
In these cases there are a few different options.
### View already has a `NSManagedObject`
Following our example above, lets say that view controller C needs to create a new object. Perhaps it is a detail view of a recipe and the user wants to add a new recipe type (perhaps she is a vegan and just discovered there is no vegan type in the list). In this case we have passed in an entity (the recipe) but not a reference to the `NSManagedObjectContext`. Fortunately this solution is easy to fix. The `NSManagedObject` retains a reference to its `NSManagedObjectContext` internally and we can access it. Therefore we can easily retrieve the `NSManagedObjectContext` from the `NSManagedObject` and create the new Type entity and pass it to the modal child or whatever our design calls for.
This again avoids rigidity because the view controller that represents the entity does not need to call up to a parent object or the `UIApplication` delegate. It is self contained and only manages view controllers that are down stream from it.
### View does not have a `NSManagedObject`
In this situation things are *slightly* more complicated. In this case we want to create a `@property` for the `NSManagedObjectContext` and require that our creator set the property.
@interface MyViewController : ViewController
@property (nonatomic, retain) NSManagedObjectContext *moc;
Again, the view controller is an island of its own because it does not care where that `NSManagedObjectContext` came from. All it knows is that it is required for the view to function. It does not care of it is a new `NSManagedObjectContext` specifically created for its use (perhaps for a cancelable edit tree) or if it is the same `NSManagedObjectContext` that has been passed around since the launch of the application. All it knows is that it has the elements it needs to perform its function.
By making the `NSManagedObjectContext` a settable property we can also transplant the view easily. If, at some point in the project lifecycle, we decide that it makes more sense to have the following design:
![Modal View Controller](http://www.cimgf.com/wp-content/uploads/2011/01/Image3.png “Modal View Controller”)
Taking from Apple’s Recipes Application, perhaps we decide that moving from the table view directly to the image of the recipe is more pleasing to the users and that when they want to see how to make it they can “flip” the image over and see the detail.
Making this change with each view controller being an island is quite simple. We just rearrange the views without having to worry too much about breaking the application.
## Solution three
Up until now we have been looking at just a navigation controller design. But what about tab bars? In the situation of a tab bar we again want to avoid rigidity because it is even more common that tabs will get moved around.
The solution to this is to again use a `@property` for the `NSManagedObjectContext` and require that the creator set this property before the view is displayed on screen. If you are creating the tabs in code this is trivial because you are already calling init on the view controller and you can add one more line of code after the init to set the property.
If the user interface is being developed mostly in Interface Builder it is slightly more tricky. Personally I am not a fan of creating navigation controllers or tab bar controllers in Interface Builder. However if that is the design then I would recommend referencing the view controllers as properties and passing along the context upon initialization of the application. It may be possible to do this entirely in Interface Builder but I am not comfortable enough to recommend that as a solution.
The overall idea behind this article is to keep each view controller separate from anything up stream or in a different silo. This will make the design far more flexible in the long run. Any time that you feel the “need” to pass in a parent view controller to a child view controller, reconsider the design. Consider using a `@protocol` delegate design or NSNotification calls instead. Keep each view controller as its own private island.
Just a quick post to point out a great article written by Nick Harris of NewsGator fame. He has looked into the issues with Core Data and encryption.
It has always been a difficult question and fortunately Apple has addressed it for us. Even better, Nick has shared the code so we don’t even need to try and discover the solution ourselves.
So Marcus is the Core Data guy, but I’ve been working with it a good bit myself lately and was recently faced with having to add re-ordering for a list of entities in a UITableView. The methods I found online for accomplishing this all suggested using an NSMutableArray as the data source for the table view. That will work, but I came up with another method, though similar, that achieved what I need without having to switch from using my NSFetchedResultsController as the data source behind the UITableView. In the end, I did use an NSMutableArray, however, I end up using it just to take advantage of its indexing. Read on to see what I mean.
An interesting question came up on Stackoverflow today so I decided to expound upon it in a short blog post.
A situation that I believe we are going to be seeing more and more often is one where application developers are writing multiple “versions” of their applications to be used on the desktop, their iPhone and now the iPad.
Because of that situation, it is becoming even more important that we write as much portable code as possible. Fortunately, our model can be completely portable between the two platforms.
Last month I was given the opportunity to write an article for The Pragmatic Programmers great magazine called “PragPub”. I am happy to say that the article I wrote for them was published in this month’s edition. The article, titled “Touching the Core”, is a walk through Apple’s great addition to the Core Data API for the iPhone.
Specifically this article walks through using the NSFetchedResultsController and some best practices in its use. The magazine is available for free on their website, [The Pragmatic Bookshelf](http://pragprog.com/magazines).
iTunes has a very neat way of searching your library, where it takes each word in your search and tries to find that word in multiple fields.Â For example, you can search for “yesterday beatles” and it will match “yesterday” in the Name field and “beatles” in the Artist field. The basic predicate binding for NSSearchField provided by Interface Builder is not complex enough to archive this kind of search.Â I need to build the predicate dynamically since I can’t assume what field the user is trying to search and that each additional word should filter the list further – just like iTunes.Â Here is how to go about adding iTunes-style searching.
A lot of hard work has gone into this book already and I see it becoming the definitive text on the subject of Core Data. The release date is slated for March 30, 2009, but it’s great to see it in beta. If you want to pick up the beta in PDF, it is available now from Pragmatic here: Core Data: Apple’s API for Persisting Data under Mac OS X.
While new Cocoa programmers will find it a great help to getting started quickly with Core Data, the book also covers some really interesting and advanced topics such as data versioning and migration, Spotlight/Quick Look integration, Sync Services, and multi-threading. You can really see Marcus’ command of the subject shine in these chapters which are already available in the beta.
Give Marcus some feedback on the book as it progresses. It’s going to be a great reference for any Cocoa Developer looking to harness the power of Core Data.
$21.00 for the beta PDF
$41.35 for the beta PDF plus hard copy when it’s released in March.
Mad props to Marcus. Congratulations!
Undo support in Cocoa is fantastic but for those who have tried to mix it with Core Data know that it can be a bit frustrating. Generally, undo support can be ignored in most applications and it will “just work”. But when Core Data is added to the recipe then things get a bit confusing and more complicated.