r209906 - Objective-C. Diagnose assigning a block pointer type to

[Argyrios Kyrtzidis]

> On May 30, 2014, at 11:34 AM, Argyrios Kyrtzidis <kyrtzidis at apple.com> wrote:
> 
> 
>> On May 30, 2014, at 11:13 AM, jahanian <fjahanian at apple.com> wrote:
>> 
>> Is this intentional, or a specific case for retain/release methods?
> 
> This also returns TRUE:
> 
> 	blk b = ^{};
> 	BOOL res = [b respondsToSelector:@selector(isKindOfClass:)];
> 
> So blocks seem to respond to NSObject protocol selectors.
> 
>> In generally though it makes no sense
>> to allow qualified id conversion of blocks pointers.
> 
> It would make sense for id<NSObject> if we would guarantee that blocks conform to NSObject protocol.

BTW, returns true as well:

	blk b = ^{};
	BOOL res = [b conformsToProtocol:@protocol(NSObject)];


> 
>> - Fariborz
>> 
>> On May 30, 2014, at 11:06 AM, Argyrios Kyrtzidis <kyrtzidis at apple.com> wrote:
>> 
>>> I’m mainly asking if blocks are implicitly implementing the NSObject protocol or not, for example this seems to work:
>>> 
>>> typedef void(^blk)(void);
>>> 
>>> int main() {
>>> 	blk b = ^{};
>>> 	BOOL res = [b respondsToSelector:@selector(retain)];
>>> 	printf("res: %d\n", res);
>>> 	return 0;
>>> }
>>> 
>>> 
>>>> On May 30, 2014, at 11:03 AM, Jordan Rose <jordan_rose at apple.com> wrote:
>>>> 
>>>> I think I agree with Fariborz on this one. I don't remember us ever promising that blocks implement the NSObject protocol, just that they are valid Objective-C objects and that they implement -retain, -release, and -copy. The spec on the Clang site doesn't even promise -retain. (http://clang.llvm.org/docs/BlockLanguageSpec.html#objective-c-extensions)
>>>> 
>>>> The developer can always cast if they really need this.
>>>> 
>>>> Jordan
>>>> 
>>>> 
>>>> On May 30, 2014, at 10:49 , jahanian <fjahanian at apple.com> wrote:
>>>> 
>>>>> It makes no sense to assign block to an object conforming to protocols.
>>>>> Such objects must implement methods of those protocols which  blocks cannot.
>>>>> I explicitly ruled this out. Did you see this in an actual user code? Curious to see how it is being used.
>>>>> 
>>>>> - Fariborz
>>>>> 
>>>>> On May 30, 2014, at 10:32 AM, Argyrios Kyrtzidis <kyrtzidis at apple.com> wrote:
>>>>> 
>>>>>> This is giving an error in the following case, is this correct ?
>>>>>> 
>>>>>> #import <Foundation/Foundation.h>
>>>>>> 
>>>>>> typedef void(^blk)(void);
>>>>>> 
>>>>>> void foo(blk b) {
>>>>>> 	id<NSObject> x = b; // error: initializing 'id<NSObject>' with an expression of incompatible type 'blk' (aka 'void (^)(void)')
>>>>>> }
>>>>>> 
>>>>> 
>>>>> _______________________________________________
>>>>> cfe-commits mailing list
>>>>> cfe-commits at cs.uiuc.edu
>>>>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>>>> 
>>> 
>> 
>