I seem to recall that in pre-C++11 standardization (probably still there though), there was a wording that mentioned something in the line of
writes should eventually be "released to memory"
Which is interesting by its vagueness, but also means that careless non-memfence/non-mutex programming still has a chance to work, with code that is multi threaded but yet (rarely) checks for some shared data.
So my question is what was/is the exact wording ?
Aucun commentaire:
Enregistrer un commentaire