2

I need to use NSInvocation to invoke a method dynamically. Here what I have tried:

NSInvocation *invocation = [NSInvocation invocationWithMethodSignature:[[messageRecord.senderController class] instanceMethodSignatureForSelector:messageRecord.receiverAction]];

[invocation setSelector:messageRecord.receiverAction];
[invocation setTarget: messageRecord.senderController];
[invocation setArgument: &(message.data) atIndex:2];
[invocation invoke];

I need to mention that, messageRecord.senderController is the object of which method will be invoked, and messageRecord.receiverAction is the selector given to this piece of code. Besides, message.data is an object of type (NSArray *) and initialized properly.

This piece of code gives the following compile-time error 

Address of property expression requested

When I change the invocation process as following, it works as expected:

NSArray *dataArray = message.data;

[invocation setSelector:messageRecord.receiverAction];
[invocation setTarget: messageRecord.senderController];
[invocation setArgument: &dataArray atIndex:2];
[invocation invoke];

The only difference between two is that: I created a local NSArray pointer and assigned message.data to it. Later, gave address of the newly created pointer instead of message.data itself.

Why it worked? What is the difference anyway?

SadullahCeran
  • 2,425
  • 4
  • 20
  • 34

2 Answers2

2

As Shimanski Ahem mentioned, a property is not something with an address. A property is just a pair of getter and setter methods. That's it. The getter and setter methods could be provided by you or synthesized by the compiler based on some instance variable. If they were provided by you, they could also be based on some instance variable, but you do some additional logic; or they could generate the value completely dynamically based on other things. Thus there might not be some "instance variable" somewhere that corresponds to the property.

If the getters and setters are synthesized, and you know the instance variable they are based on (which you can get from the synthesize declaration; which is the same as the property name if the synthesize doesn't explicitly specify one; or is _data (with an underscore in front) if the synthesize is omitted and implicitly specified in recent versions of the compiler. This is what Shimanski Artem shows.

However, step back for a moment and ask, why do you need a pointer to the value in the first place? What is the purpose? Would it make a difference if you used a pointer to the original value, or a pointer to a copy of that value? Understanding how the setArgument: method works and why it needs a pointer will help you answer many of these questions. setArgument: allows you to set one of the arguments in the invocation with a value of your choice, so it really cares about a value. So why doesn't it just have you pass the value directly? Well, what type would it be? Different types have different sizes in C, so that's not possible. So instead, it asks you to give a pointer to the value (which works for any type), and then it copies it from there. So it only cares about the value pointed to, and not the particular address. This is why (as you observed), when you want to use setArgument: with something that is not addressable (like a property, or more complicated expression, or this in Objective-C++, etc.), you can simply assign it to a temporary variable and take the address of that.

newacct
  • 119,665
  • 29
  • 163
  • 224
0

You can't take the address of a property. It's language specifics. You can try this form:

[invocation setArgument: &(message->data) atIndex:2];
Shimanski Artem
  • 1,240
  • 1
  • 11
  • 7