> On Apr 19, 2022, at 5:26 PM, Sean McBride via Cocoa-dev 
> <cocoa-dev@lists.apple.com> wrote:
> 
> On 19 Apr 2022, at 18:47, Saagar Jha wrote:
> 
>> If Thread Sanitizer says your code has a race, it almost certainly has a 
>> race.
> 
> Yeah, that's been my general experience until now.
> 
>> Your simple code seems OK superficially, but there are a couple things that 
>> could be problematic here: either your real code is actually mutating 
>> something, or (unlikely) you are touching some internal state, perhaps a CoW 
>> optimization, that is not visible to you but is silently changing things 
>> under the hood.
> 
> In case it wasn't clear, the code snippet in my email actually reproduces the 
> issue.  I created a fresh Xcode project and it's literally just:
> 
> - (void)applicationDidFinishLaunching:(NSNotification *)aNotification {
>       NSArray* array = @[@5, @6, @7, @8, @9, @10, @11];
>       
>       [array enumerateObjectsWithOptions:NSEnumerationConcurrent 
> usingBlock:^(NSNumber* num, NSUInteger idx, BOOL* stop) {
>               array[idx];
>       }];
> }     
> 
> TSan complains with this on macOS 12.3.1 on an M1 Mac Mini with Xcode 13.3.  
> But on an Intel Mac Pro, macOS 11.6.6 with Xcode 13.2.1, TSan does not 
> complain.
> 
>> In any case, I would generally suggest using -[NSArray 
>> enumerateObjectsAtIndexes:options:usingBlock:] with the 
>> NSEnumerationConcurrent flag, which should rule out any issues with 
>> concurrent access on the array itself.
> 
> I tried enumerateObjectsWithOptions:usingBlock: and TSan doesn't complain 
> with it.  But I was suspicious of that and added an NSLog to print the index, 
> and even after adding random sleeps:
> 
>       [array enumerateObjectsWithOptions:NSEnumerationConcurrent 
> usingBlock:^(NSNumber* nub, NSUInteger idx, BOOL* stop) {
>               usleep(arc4random_uniform(1000000));
>               NSLog(@"idx: %d", idx);
>               array[idx];
>       }];
> 
> The thing is *not* running concurrently on the M1.  The indices are always 
> printed in order.  Weird.
        The docs for NSEnumerationConcurrent state that it is a hint and may be 
ignored at run time.  If you absolutely need it to be run on multiple threads 
you should do it explicitly, IMO, with dispatch_async and appropriate critical 
section locking.

—Rob


_______________________________________________

Cocoa-dev mailing list (Cocoa-dev@lists.apple.com)

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 arch...@mail-archive.com

Reply via email to