Hi Jonathan, I absolutely believe that you all master this and have total knowledge of the matter. This is just a simple oversight that happens (too work,etc.): I've just pointed it out for beginners that might not catch immediately that it's a typo of the text and not a willful sentence. Gcc is great, keep up the amazing job! On Thu, Jul 20, 2023 at 3:27 PM Jonathan Wakely <jwakely.gcc@xxxxxxxxx> wrote: > On Thu, 20 Jul 2023 at 08:12, m g via Gcc-help <gcc-help@xxxxxxxxxxx> > wrote: > > > > To whom it may concern, > > > > In the example below the title "Overall Summary" the Wiki reads > > Where "the Wiki" means https://gcc.gnu.org/wiki/Atomic/GCCMM/AtomicSync > > > "*Release/acquire > > ... thread 3's assert can fail.There has been no synchronization between > > threads 1 and 3, so no value can be assumed for 'x' there.*" > > > > Are you sure about this sentence "thread 3's assert *can* fail"? > > Although the synchronization happens between threads 2 and 3; x.store > (10) > > in thread 1 still inter-thread happens before y.store (10) in thread 2 > > therefore if (y.load() == 10) is true in thread 3 thread 3's assert > > *cannot* fail. > > The statement would be true if the loads in thread 3 were relaxed. I > think you're right that for MO acquire in thread 3 there's a > happens-before relation. > > That wiki page is over a decade old and not maintained or really of > interest to anybody now. > > > > > > ( *5) A inter-thread happens-before some evaluation X, and X > inter-thread > > happens-before B* from > https://en.cppreference.com/w/cpp/atomic/memory_order > > ) > > > > *Obviously* I can be wrong, let me know! >
