Gabriel Scherer

Some further comments. You suggest to start with the same internal representation as __POS_OF__, and I think this is very reasonable. Deciding whether to hide the representation within an abstract...

I'm not worried about having 6 integers, as one could probably pack them into three Int63 values. (I'm assuming it is reasonable to assume that these line numbers or byte...

I know very little about the compilation-scheme side of weak memory models, so it is very possible that (if I understand your point correctly) the proposed model cannot be implemented...

One property that your proposal does not have, if I understand correctly, is coherence as formulated by @stedolan: if I do a relaxed read at some newer timestamp, I don't...

(If we don't update our frontier at all, I don't think we need to have a frontier at each timestamp, so I can simplify the presentation a bit: atomic locations...

> To decide between these, it might be helpful to look carefully at some intended uses of Atomic.get_relaxed to see what ordering requirements they actually have, and what the cost...

Thanks (everyone and) @polytypic for your clarifications here and the other thread, this is very helpful.

In my role as a scribe / operational semanticist, here is my attempt to translate the different models that I think are easy/natural to describe starting from the current operational...

My current understanding from your implementation-side discussions is as follows: - "older, no sync" is the weakest model, and it would probably be validated by the `!(Obj.magic a : 'a...

We discussed this at the maintainer meeting today. @stedolan clarified that his remark on reading old values later is not a criticism of the proposal itself, but rather a remark...