NSEnumerator performance vs for loop in Cocoa

问题

I know that if you have a loop that modifies the count of the items in the loop, using the NSEnumerator on a set is the best way to make sure your code blows up, however I would like to understand the performance tradeoffs between the NSEnumerator class and just an old school for loop

回答1:

Using the new for (... in ...) syntax in Objective-C 2.0 is generally the fastest way to iterate over a collection because it can maintain a buffer on the stack and get batches of items into it.

Using NSEnumerator is generally the slowest way because it often copies the collection being iterated; for immutable collections this can be cheap (equivalent to -retain) but for mutable collections it can cause an immutable copy to be created.

Doing your own iteration — for example, using -[NSArray objectAtIndex:] — will generally fall somewhere in between because while you won't have the potential copying overhead, you also won't be getting batches of objects from the underlying collection.

(PS - This question should be tagged as Objective-C, not C, since NSEnumerator is a Cocoa class and the new for (... in ...) syntax is specific to Objective-C.)

回答2:

After running the test several times, the result is almost the same. Each measure block runs 10 times consecutively.

The result in my case from the fastest to the slowest:

For..in (testPerformanceExample3) (0.006 sec)
While (testPerformanceExample4) (0.026 sec)
For(;;) (testPerformanceExample1) (0.027 sec)
Enumeration block (testPerformanceExample2) (0.067 sec)

The for and while loop is almost the same.

The tmp is an NSArray which contains 1 million objects from 0 to 999999.

- (NSArray *)createArray
{
    self.tmpArray = [NSMutableArray array];
    for (int i = 0; i < 1000000; i++)
    {
        [self.tmpArray addObject:@(i)];
    }
    return self.tmpArray;
}

The whole code:

ViewController.h

#import <UIKit/UIKit.h>

@interface ViewController : UIViewController

@property (strong, nonatomic) NSMutableArray *tmpArray;
- (NSArray *)createArray;

@end

ViewController.m

#import "ViewController.h"

@implementation ViewController

- (void)viewDidLoad {
    [super viewDidLoad];
    [self createArray];
}

- (NSArray *)createArray
{
    self.tmpArray = [NSMutableArray array];
    for (int i = 0; i < 1000000; i++)
    {
        [self.tmpArray addObject:@(i)];
    }
    return self.tmpArray;
}

@end

MyTestfile.m

#import <UIKit/UIKit.h>
#import <XCTest/XCTest.h>

#import "ViewController.h"

@interface TestCaseXcodeTests : XCTestCase
{
    ViewController *vc;
    NSArray *tmp;
}

@end

@implementation TestCaseXcodeTests

- (void)setUp {
    [super setUp];
    vc = [[ViewController alloc] init];
    tmp = vc.createArray;
}

- (void)testPerformanceExample1
{
    [self measureBlock:^{
        for (int i = 0; i < [tmp count]; i++)
        {
            [tmp objectAtIndex:i];
        }
    }];
}

- (void)testPerformanceExample2
{
    [self measureBlock:^{
        [tmp enumerateObjectsUsingBlock:^(NSNumber *obj, NSUInteger idx, BOOL *stop) {
           obj;
        }];
    }];
}

- (void)testPerformanceExample3
{
    [self measureBlock:^{
        for (NSNumber *num in tmp)
        {
            num;
        }
    }];
}

- (void)testPerformanceExample4
{
    [self measureBlock:^{
        int i = 0;
        while (i < [tmp count])
        {
            [tmp objectAtIndex:i];
            i++;
        }
    }];
}

@end

For more information visit: Apples "About Testing with Xcode"

回答3:

They are very similar. With Objective-C 2.0 most enumerations now default to NSFastEnumeration which creates a buffer of the addresses to each object in the collection that it can then deliver. The one step that you save over the classic for loop is not having to call objectAtIndex:i each time inside the loop. The internals of the collection you are enumerating implement fast enumeration with out calling objectAtIndex:i method.

The buffer is part of the reason that you can't mutate a collection as you enumerate, the address of the objects will change and the buffer that was built will no longer match.

As a bonus the format in 2.0 looks as nice as the classic for loop:

for ( Type newVariable in expression ) { 
    stmts 
}

Read the following documentaion to go deeper: NSFastEnumeration Protocol Reference

来源：https://stackoverflow.com/questions/32986/nsenumerator-performance-vs-for-loop-in-cocoa

标签

objective-c

cocoa

nsenumerator