Le 9 déc. 2011 à 08:47, Ken Thomases a écrit : > On Dec 9, 2011, at 1:11 AM, Uli Kusterer wrote: > >> On 09.12.2011, at 07:55, Ken Thomases wrote: >>> >>> Double-checked locking is broken. It is an anti-pattern in many languages, >>> including the C family under most common implementations. Don't use it. >>> Google it if you want to confirm. One reference: >>> <http://www.cs.umd.edu/~pugh/java/memoryModel/DoubleCheckedLocking.html>. >> >> Is this an issue that just exists when you want to write against ANSI C, or >> is this an actual, practical concern when writing on the Mac, for MacOS or >> iOS? > > It is an actual, practical concern. The compilers can re-order memory > accesses. The CPU can, too. > > The second and subsequent threads through the double-checked lock can see the > singleton pointer as non-nil, but the pointed-to object may not be valid. > Its content may not have been written to memory, yet. >
It was a real concern on PPC, but on Intel platform AFAIK. Memory barriers are noop on x86, as it does not perform out of order execution of memory operations. http://locklessinc.com/articles/singleton_pattern/ That said,avoiding this pattern in the future may be a good thing. -- Jean-Daniel _______________________________________________ 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: http://lists.apple.com/mailman/options/cocoa-dev/archive%40mail-archive.com This email sent to [email protected]
