On May 12, 2012, at 11:37 AM, Quincey Morris wrote:

> No, the pointer returned by 'mutableBytes' is an interior pointer. It *isn't* 
> an object pointer. (Well, as an implementation detail, it may happen to point 
> to an object's memory, but it often doesn't.)
> 
> As soon as the NSMutableData's lifetime ends -- when it reaches its dealloc 
> -- the 'mutableBytes' pointer becomes instantly invalid.
> 
> Interestingly, this appears to mean that NSData/NSMutableData has become as 
> dangerous in ARC as it is in GC. The compiler may shorten the lifetime of an 
> object variable so that the interior pointer is catastrophically invalidated. 
> This is documented in section 6.1 of the clang ARC spec:
> 
>       http://clang.llvm.org/docs/AutomaticReferenceCounting.html#optimization

Ugh, so you’re saying that ARC isn’t actually as deterministic as we’ve been 
led to believe?

> but it also give an answer:
> 
>       NSMutableData* data __attribute__ ((objc_precise_lifetime)) = …
> 
> P.S. I think there's also another, better solution, but it involves adding a 
> method to NSData/NSMutableData via a category:
> 
>       - (void*) interiorBytes __attribute__ ((objc_returns_inner_pointer)) {
>               return self.bytes;
>       }
> 
>       - (void*) mutableInteriorBytes __attribute__ 
> ((objc_returns_inner_pointer)) {
>               return self.mutableBytes;
>       }
> 
> and never using bytes/mutableBytes directly ever again. Perhaps one day 
> bytes/mutableBytes will themselves be marked this way.

On the headers on my system, bytes and mutableBytes *are* marked that way.

Charles


_______________________________________________

Cocoa-dev mailing list ([email protected])

Please do not post admin requests or moderator comments to the list.
Contact the moderators at cocoa-dev-admins(at)lists.apple.com

Help/Unsubscribe/Update your Subscription:
https://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com

This email sent to [email protected]

Reply via email to