ios - iPhone 模拟器和真实设备上关于消息转发的不同行为

Question

我想使用消息转发让任何未实现的 getter 方法返回 0，而不是抛出无法识别的选择器异常。喜欢

MyClass *r = [[MyClass alloc] init];
NSNumber *n = (NSNumber *)r;
NSLog(@"%d", [n integerValue]); // output 0
NSLog(@"%f", [n doubleValue]); // output 0.00000
NSLog(@"%@", [n stringValue]); // output (null)

所以我写了这个例子:

#pragma mark -
#pragma mark Application lifecycle

- (BOOL)application:(UIApplication *)application didFinishLaunchingWithOptions:(NSDictionary *)launchOptions {    

    NSNumber *n = (NSNumber *)self;
    NSLog(@"%d", [n integerValue]);
    NSLog(@"%f", [n doubleValue]);
    NSLog(@"%@", [n stringValue]);

    return YES;
}

- (NSMethodSignature *)methodSignatureForSelector:(SEL)aSelector {
    NSMethodSignature *ms = [super methodSignatureForSelector:aSelector];
    if(ms)
        return ms;

    // Q = uint64_t, so it should also works for double which is also 64bit
    return [NSMethodSignature signatureWithObjCTypes:"Q@:"];
}

- (void)forwardInvocation:(NSInvocation *)anInvocation {
    uint64_t ZERO64 = 0;
    [anInvocation setReturnValue:&ZERO64];
}

在真机上输出结果是0, 0.00000, (null)，在模拟器上是0, NaN, (null)

所以 double 类型没有按预期工作。我的第一个想法是将 NSMethodSignature 更改为“d@:”(d 是 double 的)

输出结果在设备和模拟器上都是正确的，但只有在模拟器上才会发生一些奇怪的事情。运行这段代码，它会在第 6 次循环时崩溃，并出现某种 CALayer 异常:

#pragma mark -
#pragma mark Application lifecycle

- (BOOL)application:(UIApplication *)application didFinishLaunchingWithOptions:(NSDictionary *)launchOptions {    

    for(NSInteger i = 0; i < 100; i++) {
        NSInteger t = [(NSNumber *)self integerValue];

        UIViewController *view = [[UIViewController alloc] init];
        // it always crash on the 6th loop on this line**
        UINavigationController *nc = [[UINavigationController alloc] initWithRootViewController:view];
    }

    return YES;
}

- (NSMethodSignature *)methodSignatureForSelector:(SEL)aSelector {
    NSMethodSignature *ms = [super methodSignatureForSelector:aSelector];
    if(ms)
        return ms;

    // we change to return double
    return [NSMethodSignature signatureWithObjCTypes:"d@:"];
}

- (void)forwardInvocation:(NSInvocation *)anInvocation {
    uint64_t ZERO64 = 0;
    [anInvocation setReturnValue:&ZERO64];
}

我对两个问题很好奇，为什么第一个示例在模拟器上返回 NaN，第二个示例发生了什么？

Answer 1

关于你的第一个问题，这是我在模拟器上发现的

union {
    double d;
    uint64_t l;
} u;
NSNumber *n = (NSNumber *)self;
u.d = [n doubleValue];
NSLog(@"%f", u.d);  // nan
NSLog(@"%llx",u.l); // fff8000000000000
bzero(&u, sizeof(double));
NSLog(@"%f", u.d);  // 0.000000
NSLog(@"%llx",u.l); // 0

很明显返回 NAN(fff8000000000000) 而不是 0.0。

要深入了解 [NSMethodSignature signatureWithObjCTypes:"d@:"] 和 [NSMethodSignature signatureWithObjCTypes:"Q@:"] 之间的区别，请看这个

NSLog(@"%@\n%@", [[NSMethodSignature signatureWithObjCTypes:"Q@:"] debugDescription], [[NSMethodSignature signatureWithObjCTypes:"d@:"] debugDescription]);

输出

<NSMethodSignature: 0x74a0950>
    number of arguments = 2
    frame size = 8
    is special struct return? NO
    return value: -------- -------- -------- --------
        type encoding (Q) 'Q'
        flags {}
        modifiers {}
        frame {offset = 0, offset adjust = 0, size = 8, size adjust = 0}
        memory {offset = 0, size = 8}
    argument 0: -------- -------- -------- --------
        type encoding (@) '@'
        flags {isObject}
        modifiers {}
        frame {offset = 0, offset adjust = 0, size = 4, size adjust = 0}
        memory {offset = 0, size = 4}
    argument 1: -------- -------- -------- --------
        type encoding (:) ':'
        flags {}
        modifiers {}
        frame {offset = 4, offset adjust = 0, size = 4, size adjust = 0}
        memory {offset = 0, size = 4}

<NSMethodSignature: 0x74a1e80>
    number of arguments = 2
    frame size = 8
    is special struct return? NO
    return value: -------- -------- -------- --------
        type encoding (d) 'd'
        flags {isFloat}    <<<<----- this flag should be set if the return value is float type
        modifiers {}
        frame {offset = 0, offset adjust = 0, size = 8, size adjust = 0}
        memory {offset = 0, size = 8}
    argument 0: -------- -------- -------- --------
        type encoding (@) '@'
        flags {isObject}
        modifiers {}
        frame {offset = 0, offset adjust = 0, size = 4, size adjust = 0}
        memory {offset = 0, size = 4}
    argument 1: -------- -------- -------- --------
        type encoding (:) ':'
        flags {}
        modifiers {}
        frame {offset = 4, offset adjust = 0, size = 4, size adjust = 0}
        memory {offset = 0, size = 4}

您可以在第二个方法签名上看到返回值上有 flags {isFloat}。我不是 x86 和 AMR 以及低级 ObjC 运行时方面的专家。但我认为CPU使用这个标志来识别返回值的类型。如果不在 x86 CPU 上设置它，预期的浮点返回值因此被解释为 NAN。

---

对于你的第二个问题，我认为这是因为你告诉运行时它将返回一个 64 位大小的值，因此堆栈上的 64 位大小的内存被归零。但是，调用者期望 32 位返回大小 (NSInteger)。因此发生了某种 stackoverflow 并导致崩溃。

---

我实际上实现了类似的东西，旨在使 NSNull 像 nil 一样工作。

- (NSMethodSignature *)methodSignatureForSelector:(SEL)aSelector {
    NSMethodSignature *signature = [super methodSignatureForSelector:aSelector];
    if (signature)
        return signature;

    const Class forwardClasses[] = {[NSNumber class], [NSString class], [NSArray class], [NSOrderedSet class]}; // add new classes if you think the list is not enough

    for (int i = 0; i < sizeof(forwardClasses)/sizeof(Class); i++) {
        Class cls = forwardClasses[i];
        signature = [cls instanceMethodSignatureForSelector:aSelector];
        if (signature) {
            return signature;
        }
    }

    return signature;
}

- (void)forwardInvocation:(NSInvocation *)anInvocation {
    NSUInteger len = [[anInvocation methodSignature] methodReturnLength];
    char buff[len];
    bzero(buff, len);
    [anInvocation setReturnValue:buff];
}