What's the Best Way to Shuffle an NSMutableArray?

If you have an NSMutableArray, how do you shuffle the elements randomly?

(I have my own answer for this, which is posted below, but I'm new to Cocoa and I'm interested to know if there is a better way.)


Update: As noted by @Mukesh, as of iOS 10+ and macOS 10.12+, there is an -[NSMutableArray shuffledArray] method that can be used to shuffle. See https://developer.apple.com/documentation/foundation/nsarray/1640855-shuffledarray?language=objc for details. (But note that this creates a new array, rather than shuffling the elements in place.)


I solved this by adding a category to NSMutableArray.

Edit: Removed unnecessary method thanks to answer by Ladd.

Edit: Changed (arc4random() % nElements) to arc4random_uniform(nElements) thanks to answer by Gregory Goltsov and comments by miho and blahdiblah

Edit: Loop improvement, thanks to comment by Ron

Edit: Added check that array is not empty, thanks to comment by Mahesh Agrawal

//  NSMutableArray_Shuffling.h

#if TARGET_OS_IPHONE
#import <UIKit/UIKit.h>
#else
#include <Cocoa/Cocoa.h>
#endif

// This category enhances NSMutableArray by providing
// methods to randomly shuffle the elements.
@interface NSMutableArray (Shuffling)
- (void)shuffle;
@end


//  NSMutableArray_Shuffling.m

#import "NSMutableArray_Shuffling.h"

@implementation NSMutableArray (Shuffling)

- (void)shuffle
{
    NSUInteger count = [self count];
    if (count <= 1) return;
    for (NSUInteger i = 0; i < count - 1; ++i) {
        NSInteger remainingCount = count - i;
        NSInteger exchangeIndex = i + arc4random_uniform((u_int32_t )remainingCount);
        [self exchangeObjectAtIndex:i withObjectAtIndex:exchangeIndex];
    }
}

@end

You don't need the swapObjectAtIndex method. exchangeObjectAtIndex:withObjectAtIndex: already exists.


Since I can't yet comment, I thought I'd contribute a full response. I modified Kristopher Johnson's implementation for my project in a number of ways (really trying to make it as concise as possible), one of them being arc4random_uniform() because it avoids modulo bias.

// NSMutableArray+Shuffling.h
#import <Foundation/Foundation.h>

/** This category enhances NSMutableArray by providing methods to randomly
 * shuffle the elements using the Fisher-Yates algorithm.
 */
@interface NSMutableArray (Shuffling)
- (void)shuffle;
@end

// NSMutableArray+Shuffling.m
#import "NSMutableArray+Shuffling.h"

@implementation NSMutableArray (Shuffling)

- (void)shuffle
{
    NSUInteger count = [self count];
    for (uint i = 0; i < count - 1; ++i)
    {
        // Select a random element between i and end of array to swap with.
        int nElements = count - i;
        int n = arc4random_uniform(nElements) + i;
        [self exchangeObjectAtIndex:i withObjectAtIndex:n];
    }
}

@end

If you import GameplayKit, there is a shuffled API:

https://developer.apple.com/reference/foundation/nsarray/1640855-shuffled

let shuffledArray = array.shuffled()

A slightly improved and concise solution (compared to the top answers).

The algorithm is the same and is described in literature as "Fisher-Yates shuffle".

In Objective-C:

@implementation NSMutableArray (Shuffle)
// Fisher-Yates shuffle
- (void)shuffle
{
    for (NSUInteger i = self.count; i > 1; i--)
        [self exchangeObjectAtIndex:i - 1 withObjectAtIndex:arc4random_uniform((u_int32_t)i)];
}
@end

In Swift 3.2 and 4.x:

extension Array {
    /// Fisher-Yates shuffle
    mutating func shuffle() {
        for i in stride(from: count - 1, to: 0, by: -1) {
            swapAt(i, Int(arc4random_uniform(UInt32(i + 1))))
        }
    }
}

In Swift 3.0 and 3.1:

extension Array {
    /// Fisher-Yates shuffle
    mutating func shuffle() {
        for i in stride(from: count - 1, to: 0, by: -1) {
            let j = Int(arc4random_uniform(UInt32(i + 1)))
            (self[i], self[j]) = (self[j], self[i])
        }
    }
}

Note: A more concise solution in Swift is possible from iOS10 using GameplayKit.

Note: An algorithm for unstable shuffling (with all positions forced to change if count > 1) is also available