@pkhuong@wingo (Paul can correct me but I'm pretty sure even the main experts who designed the C++ memory model now consider memory_order_consume to have been a mistake.)
@pervognsen@pkhuong@wingo mo_consume was a mistake but data dependencies are good and there was a proposal for how to do it right for c++ which is still in limbo....
@moonchild@pkhuong@pervognsen i have a general theory that people that conclude they can use a memory order that is not acq/rel or seq_cst are smarter than me but more foolish
@wingo@moonchild@pkhuong There are plenty of cases with mo_relaxed which are easy to think about but in the general case it's definitely nebulous. I assume you've seen all the out-of-thin-air (OOTA) stuff with mo_relaxed... Apparently mo_relaxed, as it stands, didn't formalize what the designers thought it formalized. It's kind of like the Peano axioms and the standard model of natural numbers vs non-standard models.