哪种遍历NSArray/NSDictionary方式快？——一个容器遍历测试

哪种遍历NSArray/NSDictionary方式快？——一个容器遍历测试   

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。