Behind the Scenes of Objective-C 2.0

• By David Chisnall
• Nov 3, 2006
• Article is provided courtesy of Prentice Hall Professional

Like this article? We recommend
Programming in Objective-C

The Future

Apple is traditionally very tight-lipped about future products. For example, the Leopard preview site has very few details on the new features added to Objective-C. In recent years, however, Apple has tried hard to "play nice" with the Free Software community—in particular, by minimizing the divergence of the Apple version of the GCC from the main branch. At any point, however, a number of patches haven’t made it into the main tree. For this reason, Apple has its own branch in the GNU CVS repository.

Since Apple’s GCC development branch is public, it’s possible to poke around in it and see what the Apple developers are up to. This isn’t an endeavor for the faint-hearted; the GCC codebase is not the easiest to understand at the best of times. Wouldn’t it be easier if there were a set of simple example programs showing the new features? Well, I closed my eyes and wished hard, and the magical code pixies delivered this link for the Objective-C test suite in the Apple branch. Every new feature has a simple example program used to test functionality. All you need to do is compile them.

If only life were that simple.

While Apple’s Objective-C compiler is Free Software (due to being a branch of the GNU compiler), Apple’s Objective-C runtime library is not. Not only is it not Free Software—the version required by Objective-C 2.0 isn’t released yet. While it’s possible to compile these programs, some of them won’t link, and most of them won’t work. Still, it’s possible to get an idea of what they do, even without compiling.

Garbage Collection

Pretty much any Objective-C programmer will tell you that the one feature he really misses when using Objective-C is garbage collection. OpenStep uses a retain/release reference-counting mechanism that works most of the time, but is sometimes difficult to use (for example, with data structures containing loops).

Garbage collection in Objective-C isn’t exactly new. Tiger shipped with a compiler supporting garbage collection, and the main GNU trunk has supported the Boehm Garbage Collector for a while. The problem is largely with the standard libraries, which must be modified to support garbage collection.

Objective-C 2.0 adds __strong and __weak type qualifiers for pointers. These qualifiers control whether the compiler should notify the runtime garbage collector of assignments to pointers. A pointer defined at __strong will cause garbage collection events to be generated for all events. The __weak qualifier is redundant in declarations (__weak is the default), but it can be used to suppress the generation of garbage collection events by casting a pointer to a __weak pointer before assignment.

Examples of the use of these features can be found in the objc-gc-*.m test cases.

Loose Protocol Definitions

A protocol is a strict formal definition of an interface to a class. Any class that adopts a protocol must implement it. But sometimes you don’t need all of the methods in an interface. You may want to specify a core set of methods as well as a set of convenience methods that might allow more efficiency, but aren’t absolutely required.

The new @optional and @required keywords permit this flexibility. Any method in a protocol defined as @required behaves like a method in a traditional Objective-C protocol; all classes that adopt the protocol must implement it. @optional methods are advisory; they’re ones that a class can implement, but doesn’t have to. A string class, for example, might be required to provide methods for inserting and removing objects at a specific location, but may also provide a method for replacing a character, which would be more efficient than inserting and deleting as two separate operations.

Examples of the use of these keywords can be found in the enhanced-proto-*.m test cases.

Concrete Protocols

Objective-C, just like Smalltalk, doesn’t have multiple inheritance. This was a design decision, as it was with Java, because multiple inheritance can cause headaches. Instead, Objective-C has categories, which allow methods to be added to an existing class, and protocols that specify interfaces.

Objective-C 2.0 adds the concept of a concrete protocol. Whereas a protocol is a definition of an interface, a concrete protocol adds some implementation detail as well. If the class that adopts the protocol doesn’t provide its own implementation, it will get the default implementation from the protocol.

These features can be combined effectively with @optional methods in a protocol to provide a default implementation of @optional methods, but allow an implementer to replace them with a more useful method.

Method Attributes

A while ago, the GCC added function attributes to C. These are specified using __attribute__ in the function declaration, and provide some extra metadata to the compiler. The uses of function attributes can be divided into two cases:

• Optimization hints tell the compiler to do things such as pass some parameters in registers, or reuse results when the same function is called with the same arguments.
• Coding hints are used to generate warnings when a function is invoked.

Most of the optimization hints don’t make sense in the context of Objective-C methods, since they’re invoked via the objc_msgSend function. The coding hints do make sense, however. Objective-C 2.0 adds support for deprecated and unavailable attributes. If a program calls methods marked with either of these descriptions, the compiler will emit a warning suggesting that the method not be used.

Examples of these options can be found in the method-attribute-*.m test cases.

Properties

One nice feature of recent versions of Objective-C is key-value coding, which provides an abstracted interface to getting and setting instance variables. Instance variables are accessed via the objectForKey: method, which calls the correct get method, returns the instance variable, or calls a default handler.

Objective-C 2.0 takes this arrangement a step further with properties. Conceptually similar to public instance variables, properties are more clever. The declaration of a property looks like this:

@property int aProperty;

This syntax provides nice encapsulation, since it gives no information about how the property is accessed. In the implementation section, you might have something like this:

@property (ivar = anInstanceVariable, copies, setter = aSetMethod:) int aProperty;

This would (I believe) return a copy of anInstanceVariable when someone tried to read the property, and invoke the aSetMethod method when it was set. Properties are accessed using the following syntax:

anObject.aProperty = 2;

This is a new syntactic addition to Objective-C objects. Instance variables are accessed using ->, since objects are always handled via pointers. However, it’s consistent with the separator used in key-value coding paths.

Note that property names must be distinct from instance variable names, or an error will be generated.

The test cases related to properties are named property-*.m.

For Each

When you want to iterate over the elements in a collection in Objective-C, you typically create an enumerator and use that. Because this is something that’s commonly done, I wrote a macro to reduce the number of copy-and-paste errors:

#define FOREACH(collection,object,type) FOREACHE(collection,object,type,object ## enumerator)
#define FOREACHE(collection,object,type,enumerator)NSEnumerator * enumerator = [collection objectEnumerator];type object;IMP next ## object ## in ## enumerator = [enumerator methodForSelector:@selector(nextObject)];while((object = next ## object ## in ## enumerator(enumerator,@selector(nextObject))))

This works quite well, and gains a little bit of speed by caching the instance method pointer for the nextObject method in the enumerator. To log every string in an array, you would use it as follows:

FOREACH(anArray, string, NSString*)
{
  NSLog(@"%@", string);
}

Objective-C 2.0 adds a new for statement construct for doing this. Here’s the equivalent code in Objective-C 2.0:

for(NSString * string in anArray)
{
  NSLog(@"%@", string);
}

The most interesting thing about this is that it doesn’t work on any of the standard collections without modification. It depends on the existence of a method with the following signature:

- (unsigned int)countByEnumeratingWithState:(struct __objcFastEnumerationState *)state objects:(id *)items count:(unsigned int)stackcount;

It’s not immediately obvious why Apple chose this path, since the existing enumerator framework allows this functionality without modification to existing code. Inspecting the __objcFastEnumerationState structure doesn’t provide much more insight, so I poked a little deeper. For reference, the code related to handling the for each statement is found in gcc/objc/objc-act.c.

The for each loop invokes this method to get a C array of up to 16 elements at a time. These are returned via items, and the return value of the method contains the number of elements in the array. All quite simple so far—and a significant efficiency increase, since you need only one method call per 16 elements. In the simple case, the following are roughly equivalent:

for(type element in collection)
{
  //Do Something
}

and

type element;
struct __objcFastEnumerationState state = { 0 };
id items[16];
unsigned long limit
while((limit = [collection countByEnumeratingWithState:&state
            objects:items
             count:16]))
{
  for(unsigned int i=0 ; i<limit ; i++)
  {
    element = items[i];
    //Do Something
  }
}

This is not quite accurate, though. In cases where the data is stored internally in a C array, the object can return a pointer to that array in state.itemsPtr to avoid copying. In other cases, it sets state.itemsPtr to items and copies the requisite pointers into items.

Tantalizing, however, is the fact that the objc_enumerationMutation() is called in cases where subsequent calls to countByEnumeratingWithState:objects:count change the value of the long integer pointed to by state.mutationsPtr. Unfortunately, objc_enumerationMutation() is a runtime library function that doesn’t exist in published versions of the library, so it’s unclear exactly what it does.