哪种遍历NSArray/NSDictionary方式快？

做app的时候，总免不了要多次遍历数组或者字典。

究竟哪种遍历方式比较快呢？我做了如下测试：

首先定义测试用宏：

#define MULogTimeintervalBegin(INFO) NSTimeInterval start = [NSDate timeIntervalSinceReferenceDate];\
							NSTimeInterval duration = 0;\
							NSLog(@"MULogTimeintervalBegin:%@", INFO)

#define MULogTimeintervalPauseAndLog(INFO) duration = [NSDate timeIntervalSinceReferenceDate] - start;\
							start += duration;\
							NSLog(@"%@:%f", INFO, duration);\
							duration = 0

#define TESTSCALE 100000

接着编写测试代码：

NSarray：

- (void)testArray
{
    NSMutableArray* testArray = [NSMutableArray arrayWithCapacity:TESTSCALE];
    for (NSInteger i = 1; i <= TESTSCALE; ++i) {
        [testArray addObject:[NSString stringWithFormat:@"%ld", i]];
    }
    NSLog(@"init:%ld", [testArray count]);
    
    __block NSMutableString* sum = [NSMutableString stringWithCapacity:TESTSCALE];
    
    MULogTimeintervalBegin(@"ArrayTest");
    NSUInteger count = [testArray count];
    for (NSInteger i = 0; i < count; ++i) {
        [sum appendString:[testArray objectAtIndex:i]];
    }
    [sum setString:@""];
    MULogTimeintervalPauseAndLog(@"for statement");
    
    for(NSString* item in testArray) {
        [sum appendString:item];
    }
    [sum setString:@""];
    MULogTimeintervalPauseAndLog(@"for-in");
    
    [testArray enumerateObjectsUsingBlock:^(id obj, NSUInteger idx, BOOL *stop) {
        [sum appendString:obj];
    }];
    [sum setString:@""];
    MULogTimeintervalPauseAndLog(@"enumerateBlock");
}

NSDictionary：

- (void)testDictionary {
    NSMutableDictionary* testDic = [NSMutableDictionary dictionaryWithCapacity:TESTSCALE];
    for (NSInteger i = 1; i <= TESTSCALE; ++i) {
        [testDic setObject:@"test" forKey:[NSString stringWithFormat:@"%ld", i]];
    }
    NSLog(@"init:%ld", [testDic count]);
    
    __block NSMutableString* sum = [NSMutableString stringWithCapacity:TESTSCALE];
    
    MULogTimeintervalBegin(@"DictionaryTest");
    for (NSString* object in [testDic allValues]) {
        [sum appendString:object];
    }
    [sum setString:@""];
    MULogTimeintervalPauseAndLog(@"for statement allValues");
    
    for (id akey in [testDic allKeys]) {
        [sum appendString:[testDic objectForKey:akey]];
    }
    [sum setString:@""];
    MULogTimeintervalPauseAndLog(@"for statement allKeys");
    
    [testDic enumerateKeysAndObjectsUsingBlock:^(id key, id obj, BOOL *stop) {
        [sum appendString:obj];
    } ];
    MULogTimeintervalPauseAndLog(@"enumeration");
}

下面是测试结果：

Test Case '-[LoopTestTests testArray]' started.

2012-08-02 17:14:22.061 otest[388:303] init:100000

2012-08-02 17:14:22.062 otest[388:303] MULogTimeintervalBegin:ArrayTest

2012-08-02 17:14:22.075 otest[388:303]for statement:0.013108

2012-08-02 17:14:22.083 otest[388:303]for-in:0.008186

2012-08-02 17:14:22.095 otest[388:303]enumerateBlock:0.012290

Test Case '-[LoopTestTests testArray]' passed (0.165 seconds).

Test Case '-[LoopTestTests testDictionary]' started.

2012-08-02 17:14:22.273 otest[388:303] init:100000

2012-08-02 17:14:22.274 otest[388:303] MULogTimeintervalBegin:DictionaryTest

2012-08-02 17:14:22.284 otest[388:303]for statement allValues:0.010566

2012-08-02 17:14:22.307 otest[388:303]for statement allKeys:0.022377

2012-08-02 17:14:22.330 otest[388:303]enumeration:0.023914

Test Case '-[LoopTestTests testDictionary]' passed (0.217 seconds).

可以看出对于数组来说，for-in方式遍历速度是最快的，普通风格的for和block方式速度差不多。对于字典来说，allValues方式遍历最快，allKeys和block差不多。

那么，为什么会这样呢？

NSArray：

for (NSInteger i = 0; i < count; ++i) {
        [sum appendString:[testArray objectAtIndex:i]];
    }

这里由于存在：[objectAtIndex:i]这样的取操作，所以速度会有所下降。

而

for(NSString* item in testArray) {
        [sum appendString:item];
    }

尽管也有取操作，但是绕开了oc的message机制，速度会快一点。也有可能是编译器为了for-in作了优化。

block为什么会慢一些这个有待研究。

NSDictionary：

for (id akey in [testDic allKeys]) {
        [sum appendString:[testDic objectForKey:akey]];
    }

这个就很明显了，第二种方法多了一次objectForKey的操作。block的话有待研究。

=================================================================================

google了一下，stackoverflow上面有类似的讨论：点击打开链接

大意是：for-in语法会对容器里面的元素的内存地址建立一个缓冲，遍历的时候从缓冲直接取得元素的地址而不是通过调用方法来获取，所以效率比较高。另外，这也是不能在循环体中修改容器元素的原因之一。