[objective-c] What should my Objective-C singleton look like?

My singleton accessor method is usually some variant of:

static MyClass *gInstance = NULL;

+ (MyClass *)instance
{
    @synchronized(self)
    {
        if (gInstance == NULL)
            gInstance = [[self alloc] init];
    }

    return(gInstance);
}

What could I be doing to improve this?

My way is simple like this:

static id instanceOfXXX = nil;

+ (id) sharedXXX
{
    static volatile BOOL initialized = NO;

    if (!initialized)
    {
        @synchronized([XXX class])
        {
            if (!initialized)
            {
                instanceOfXXX = [[XXX alloc] init];
                initialized = YES;
            }
        }
    }

    return instanceOfXXX;
}

If the singleton is initialized already, the LOCK block will not be entered. The second check if(!initialized) is to make sure it is not initialized yet when the current thread acquires the LOCK.

@interface MySingleton : NSObject
{
}

+ (MySingleton *)sharedSingleton;
@end

@implementation MySingleton

+ (MySingleton *)sharedSingleton
{
  static MySingleton *sharedSingleton;

  @synchronized(self)
  {
    if (!sharedSingleton)
      sharedSingleton = [[MySingleton alloc] init];

    return sharedSingleton;
  }
}

@end

[Source]

The accepted answer, although it compiles, is incorrect.

+ (MySingleton*)sharedInstance
{
    @synchronized(self)  <-------- self does not exist at class scope
    {
        if (sharedInstance == nil)
            sharedInstance = [[MySingleton alloc] init];
    }
    return sharedInstance;
}

Per Apple documentation:

... You can take a similar approach to synchronize the class methods of the associated class, using the Class object instead of self.

Even if using self works, it shouldn't and this looks like a copy and paste mistake to me. The correct implementation for a class factory method would be:

+ (MySingleton*)getInstance
{
    @synchronized([MySingleton class]) 
    {
        if (sharedInstance == nil)
            sharedInstance = [[MySingleton alloc] init];
    }
    return sharedInstance;
}

For an in-depth discussion of the singleton pattern in Objective-C, look here:

Using the Singleton Pattern in Objective-C

To extend the example from @robbie-hanson ...

static MySingleton* sharedSingleton = nil;

+ (void)initialize {
    static BOOL initialized = NO;
    if (!initialized) {
        initialized = YES;
        sharedSingleton = [[self alloc] init];
    }
}

- (id)init {
    self = [super init];
    if (self) {
        // Member initialization here.
    }
    return self;
}

static mySingleton *obj=nil;

@implementation mySingleton

-(id) init {
    if(obj != nil){     
        [self release];
        return obj;
    } else if(self = [super init]) {
        obj = self;
    }   
    return obj;
}

+(mySingleton*) getSharedInstance {
    @synchronized(self){
        if(obj == nil) {
            obj = [[mySingleton alloc] init];
        }
    }
    return obj;
}

- (id)retain {
    return self;
}

- (id)copy {
    return self;
}

- (unsigned)retainCount {
    return UINT_MAX;  // denotes an object that cannot be released
}

- (void)release {
    if(obj != self){
        [super release];
    }
    //do nothing
}

- (id)autorelease {
    return self;
}

-(void) dealloc {
    [super dealloc];
}
@end

I've rolled singleton into a class, so other classes can inherit singleton properties.

Singleton.h :

static id sharedInstance = nil;

#define DEFINE_SHARED_INSTANCE + (id) sharedInstance {  return [self sharedInstance:&sharedInstance]; } \
                               + (id) allocWithZone:(NSZone *)zone { return [self allocWithZone:zone forInstance:&sharedInstance]; }

@interface Singleton : NSObject {

}

+ (id) sharedInstance;
+ (id) sharedInstance:(id*)inst;

+ (id) allocWithZone:(NSZone *)zone forInstance:(id*)inst;

@end

Singleton.m :

#import "Singleton.h"


@implementation Singleton


+ (id) sharedInstance { 
    return [self sharedInstance:&sharedInstance];
}

+ (id) sharedInstance:(id*)inst {
    @synchronized(self)
    {
        if (*inst == nil)
            *inst = [[self alloc] init];
    }
    return *inst;
}

+ (id) allocWithZone:(NSZone *)zone forInstance:(id*)inst {
    @synchronized(self) {
        if (*inst == nil) {
            *inst = [super allocWithZone:zone];
            return *inst;  // assignment and return on first allocation
        }
    }
    return nil; // on subsequent allocation attempts return nil
}

- (id)copyWithZone:(NSZone *)zone {
    return self;
}

- (id)retain {
    return self;
}

- (unsigned)retainCount {
    return UINT_MAX;  // denotes an object that cannot be released
}

- (void)release {
    //do nothing
}

- (id)autorelease {
    return self;
}


@end

And here is an example of some class, that you want to become singleton.

#import "Singleton.h"

@interface SomeClass : Singleton {

}

@end

@implementation SomeClass 

DEFINE_SHARED_INSTANCE;

@end

The only limitation about Singleton class, is that it is NSObject subclass. But most time I use singletons in my code they are in fact NSObject subclasses, so this class really ease my life and make code cleaner.

A thorough explanation of the Singleton macro code is on the blog Cocoa With Love

http://cocoawithlove.com/2008/11/singletons-appdelegates-and-top-level.html.

I know there are a lot of comments on this "question", but I don't see many people suggesting using a macro to define the singleton. It's such a common pattern and a macro greatly simplifies the singleton.

Here are the macros I wrote based on several Objc implementations I've seen.

Singeton.h

/**
 @abstract  Helps define the interface of a singleton.
 @param  TYPE  The type of this singleton.
 @param  NAME  The name of the singleton accessor.  Must match the name used in the implementation.
 @discussion
 Typcially the NAME is something like 'sharedThing' where 'Thing' is the prefix-removed type name of the class.
 */
#define SingletonInterface(TYPE, NAME) \
+ (TYPE *)NAME;


/**
 @abstract  Helps define the implementation of a singleton.
 @param  TYPE  The type of this singleton.
 @param  NAME  The name of the singleton accessor.  Must match the name used in the interface.
 @discussion
 Typcially the NAME is something like 'sharedThing' where 'Thing' is the prefix-removed type name of the class.
 */
#define SingletonImplementation(TYPE, NAME) \
static TYPE *__ ## NAME; \
\
\
+ (void)initialize \
{ \
    static BOOL initialized = NO; \
    if(!initialized) \
    { \
        initialized = YES; \
        __ ## NAME = [[TYPE alloc] init]; \
    } \
} \
\
\
+ (TYPE *)NAME \
{ \
    return __ ## NAME; \
}

Example of use:

MyManager.h

@interface MyManager

SingletonInterface(MyManager, sharedManager);

// ...

@end

MyManager.m

@implementation MyManager

- (id)init
{
    self = [super init];
    if (self) {
        // Initialization code here.
    }

    return self;
}

SingletonImplementation(MyManager, sharedManager);

// ...

@end

Why a interface macro when it's nearly empty? Code consistency between the header and code files; maintainability in case you want to add more automatic methods or change it around.

I'm using the initialize method to create the singleton as is used in the most popular answer here (at time of writing).

Since Kendall posted a threadsafe singleton that attempts to avoid locking costs, I thought I would toss one up as well:

#import <libkern/OSAtomic.h>

static void * volatile sharedInstance = nil;                                                

+ (className *) sharedInstance {                                                                    
  while (!sharedInstance) {                                                                          
    className *temp = [[self alloc] init];                                                                 
    if(!OSAtomicCompareAndSwapPtrBarrier(0x0, temp, &sharedInstance)) {
      [temp release];                                                                                   
    }                                                                                                    
  }                                                                                                        
  return sharedInstance;                                                                        
}

Okay, let me explain how this works:

1. Fast case: In normal execution sharedInstance has already been set, so the while loop is never executed and the function returns after simply testing for the variable's existence;

2. Slow case: If sharedInstance doesn't exist, then an instance is allocated and copied into it using a Compare And Swap ('CAS');

3. Contended case: If two threads both attempt to call sharedInstance at the same time AND sharedInstance doesn't exist at the same time then they will both initialize new instances of the singleton and attempt to CAS it into position. Whichever one wins the CAS returns immediately, whichever one loses releases the instance it just allocated and returns the (now set) sharedInstance. The single OSAtomicCompareAndSwapPtrBarrier acts as both a write barrier for the setting thread and a read barrier from the testing thread.

With Objective C class methods, we can just avoid using the singleton pattern the usual way, from:

[[Librarian sharedInstance] openLibrary]

to:

[Librarian openLibrary]

by wrapping the class inside another class that just has Class Methods, that way there is no chance of accidentally creating duplicate instances, as we're not creating any instance!

I wrote a more detailed blog here :)

Just wanted to leave this here so I don't lose it. The advantage to this one is that it's usable in InterfaceBuilder, which is a HUGE advantage. This is taken from another question that I asked:

static Server *instance;

+ (Server *)instance { return instance; }

+ (id)hiddenAlloc
{
    return [super alloc];
}

+ (id)alloc
{
    return [[self instance] retain];
}


+ (void)initialize
{
    static BOOL initialized = NO;
    if(!initialized)
    {
        initialized = YES;
        instance = [[Server hiddenAlloc] init];
    }
}

- (id) init
{
    if (instance)
        return self;
    self = [super init];
    if (self != nil) {
        // whatever
    }
    return self;
}

static MyClass *sharedInst = nil;

+ (id)sharedInstance
{
    @synchronize( self ) {
        if ( sharedInst == nil ) {
            /* sharedInst set up in init */
            [[self alloc] init];
        }
    }
    return sharedInst;
}

- (id)init
{
    if ( sharedInst != nil ) {
        [NSException raise:NSInternalInconsistencyException
            format:@"[%@ %@] cannot be called; use +[%@ %@] instead"],
            NSStringFromClass([self class]), NSStringFromSelector(_cmd), 
            NSStringFromClass([self class]),
            NSStringFromSelector(@selector(sharedInstance)"];
    } else if ( self = [super init] ) {
        sharedInst = self;
        /* Whatever class specific here */
    }
    return sharedInst;
}

/* These probably do nothing in
   a GC app.  Keeps singleton
   as an actual singleton in a
   non CG app
*/
- (NSUInteger)retainCount
{
    return NSUIntegerMax;
}

- (oneway void)release
{
}

- (id)retain
{
    return sharedInst;
}

- (id)autorelease
{
    return sharedInst;
}

Edit: This implementation obsoleted with ARC. Please have a look at How do I implement an Objective-C singleton that is compatible with ARC? for correct implementation.

All the implementations of initialize I've read in other answers share a common error.

+ (void) initialize {
  _instance = [[MySingletonClass alloc] init] // <----- Wrong!
}

+ (void) initialize {
  if (self == [MySingletonClass class]){ // <----- Correct!
      _instance = [[MySingletonClass alloc] init] 
  }
}

The Apple documentation recommend you check the class type in your initialize block. Because subclasses call the initialize by default. There exists a non-obvious case where subclasses may be created indirectly through KVO. For if you add the following line in another class:

[[MySingletonClass getInstance] addObserver:self forKeyPath:@"foo" options:0 context:nil]

Objective-C will implicitly create a subclass of MySingletonClass resulting in a second triggering of +initialize.

You may think that you should implicitly check for duplicate initialization in your init block as such:

- (id) init { <----- Wrong!
   if (_instance != nil) {
      // Some hack
   }
   else {
      // Do stuff
   }
  return self;
}

But you will shoot yourself in the foot; or worse give another developer the opportunity to shoot themselves in the foot.

- (id) init { <----- Correct!
   NSAssert(_instance == nil, @"Duplication initialization of singleton");
   self = [super init];
   if (self){
      // Do stuff
   }
   return self;
}

TL;DR, here's my implementation

@implementation MySingletonClass
static MySingletonClass * _instance;
+ (void) initialize {
   if (self == [MySingletonClass class]){
      _instance = [[MySingletonClass alloc] init];
   }
}

- (id) init {
   ZAssert (_instance == nil, @"Duplication initialization of singleton");
   self = [super init];
   if (self) {
      // Initialization
   }
   return self;
}

+ (id) getInstance {
   return _instance;
}
@end

(Replace ZAssert with our own assertion macro; or just NSAssert.)

Here's a macro that I put together:

http://github.com/cjhanson/Objective-C-Optimized-Singleton

It is based on the work here by Matt Gallagher But changing the implementation to use method swizzling as described here by Dave MacLachlan of Google.

I welcome comments / contributions.

You don't want to synchronize on self... Since the self object doesn't exist yet! You end up locking on a temporary id value. You want to ensure that no one else can run class methods ( sharedInstance, alloc, allocWithZone:, etc ), so you need to synchronize on the class object instead:

@implementation MYSingleton

static MYSingleton * sharedInstance = nil;

+( id )sharedInstance {
    @synchronized( [ MYSingleton class ] ) {
        if( sharedInstance == nil )
            sharedInstance = [ [ MYSingleton alloc ] init ];
    }

    return sharedInstance;
}

+( id )allocWithZone:( NSZone * )zone {
    @synchronized( [ MYSingleton class ] ) {
        if( sharedInstance == nil )
            sharedInstance = [ super allocWithZone:zone ];
    }

    return sharedInstance;
}

-( id )init {
    @synchronized( [ MYSingleton class ] ) {
        self = [ super init ];
        if( self != nil ) {
            // Insert initialization code here
        }

        return self;
    }
}

@end

This works in a non-garbage collected environment also.

@interface MySingleton : NSObject {
}

+(MySingleton *)sharedManager;

@end


@implementation MySingleton

static MySingleton *sharedMySingleton = nil;

+(MySingleton*)sharedManager {
    @synchronized(self) {
        if (sharedMySingleton == nil) {
            [[self alloc] init]; // assignment not done here
        }
    }
    return sharedMySingleton;
}


+(id)allocWithZone:(NSZone *)zone {
    @synchronized(self) {
        if (sharedMySingleton == nil) {
            sharedMySingleton = [super allocWithZone:zone];
            return sharedMySingleton;  // assignment and return on first allocation
        }
    }
    return nil; //on subsequent allocation attempts return nil
}


-(void)dealloc {
    [super dealloc];
}

-(id)copyWithZone:(NSZone *)zone {
    return self;
}


-(id)retain {
    return self;
}


-(unsigned)retainCount {
    return UINT_MAX;  //denotes an object that cannot be release
}


-(void)release {
    //do nothing    
}


-(id)autorelease {
    return self;    
}


-(id)init {
    self = [super init];
    sharedMySingleton = self;

    //initialize here

    return self;
}

@end

I usually use code roughly similar to that in Ben Hoffstein's answer (which I also got out of Wikipedia). I use it for the reasons stated by Chris Hanson in his comment.

However, sometimes I have a need to place a singleton into a NIB, and in that case I use the following:

@implementation Singleton

static Singleton *singleton = nil;

- (id)init {
    static BOOL initialized = NO;
    if (!initialized) {
        self = [super init];
        singleton = self;
        initialized = YES;
    }
    return self;
}

+ (id)allocWithZone:(NSZone*)zone {
    @synchronized (self) {
        if (!singleton)
            singleton = [super allocWithZone:zone];     
    }
    return singleton;
}

+ (Singleton*)sharedSingleton {
    if (!singleton)
        [[Singleton alloc] init];
    return singleton;
}

@end

I leave the implementation of -retain (etc.) to the reader, although the above code is all you need in a garbage collected environment.

How about

static MyClass *gInstance = NULL;

+ (MyClass *)instance
{
    if (gInstance == NULL) {
        @synchronized(self)
        {
            if (gInstance == NULL)
                gInstance = [[self alloc] init];
        }
    }

    return(gInstance);
}

So you avoid the synchronization cost after initialization?

I've not read through all the solutions, so forgive if this code is redundant.

This is the most thread safe implementation in my opinion.

+(SingletonObject *) sharedManager
{
    static SingletonObject * sharedResourcesObj = nil;

    @synchronized(self)
    {
        if (!sharedResourcesObj)
        {
            sharedResourcesObj = [[SingletonObject alloc] init];
        }
    }

    return sharedResourcesObj;
}

Short answer: Fabulous.

Long answer: Something like....

static SomeSingleton *instance = NULL;

@implementation SomeSingleton

+ (id) instance {
    static dispatch_once_t onceToken;
    dispatch_once(&onceToken, ^{
        if (instance == NULL){
            instance = [[super allocWithZone:NULL] init];
        }
    });
    return instance;
}

+ (id) allocWithZone:(NSZone *)paramZone {
    return [[self instance] retain];
}

- (id) copyWithZone:(NSZone *)paramZone {
    return self;
}

- (id) autorelease {
    return self;
}

- (NSUInteger) retainCount {
    return NSUIntegerMax;
}

- (id) retain {
    return self;
}

@end

Be sure to read the dispatch/once.h header to understand what's going on. In this case the header comments are more applicable than the docs or man page.

Shouln't this be threadsafe and avoid the expensive locking after the first call?

+ (MySingleton*)sharedInstance
{
    if (sharedInstance == nil) {
        @synchronized(self) {
            if (sharedInstance == nil) {
                sharedInstance = [[MySingleton alloc] init];
            }
        }
    }
    return (MySingleton *)sharedInstance;
}

KLSingleton is:

1. Subclassible (to the n-th degree)
2. ARC compatible
3. Safe with alloc and init
4. Loaded lazily
5. Thread-safe
6. Lock-free (uses +initialize, not @synchronize)
7. Macro-free
8. Swizzle-free
9. Simple

KLSingleton

Per my other answer below, I think you should be doing:

+ (id)sharedFoo
{
    static dispatch_once_t once;
    static MyFoo *sharedFoo;
    dispatch_once(&once, ^ { sharedFoo = [[self alloc] init]; });
    return sharedFoo;
}

I have an interesting variation on sharedInstance that is thread safe, but does not lock after the initialization. I am not yet sure enough of it to modify the top answer as requested, but I present it for further discussion:

// Volatile to make sure we are not foiled by CPU caches
static volatile ALBackendRequestManager *sharedInstance;

// There's no need to call this directly, as method swizzling in sharedInstance
// means this will get called after the singleton is initialized.
+ (MySingleton *)simpleSharedInstance
{
    return (MySingleton *)sharedInstance;
}

+ (MySingleton*)sharedInstance
{
    @synchronized(self)
    {
        if (sharedInstance == nil)
        {
            sharedInstance = [[MySingleton alloc] init];
            // Replace expensive thread-safe method 
            // with the simpler one that just returns the allocated instance.
            SEL origSel = @selector(sharedInstance);
            SEL newSel = @selector(simpleSharedInstance);
            Method origMethod = class_getClassMethod(self, origSel);
            Method newMethod = class_getClassMethod(self, newSel);
            method_exchangeImplementations(origMethod, newMethod);
        }
    }
    return (MySingleton *)sharedInstance;
}