[atomics.types.operations]p20's note makes incorrect assumptions about atomic representation

[zygoloid]

[ Note: For example, the effect of compare_exchange_strong is

if (memcmp(this, &expected, sizeof(*this)) == 0)
  memcpy(this, &desired, sizeof(*this));
else
  memcpy(expected, this, sizeof(*this));

— end note ]

This is wrong in a multitude of ways... not least of which is that that's not even atomic. But also this assumes that sizeof(atomic<T>) == sizeof(T), that the layout of an atomic directly contains a T, and so on.

[jensmaurer]

Do we want to deal with this editorially, or have SG1 chime in?

[jwakely]

IIRC SG1 came up with that example, so I think getting their input would be best.

[jfbastien]

I can whip up a paper that fixes it if you think that's helpful? Another approach would be to wait for us to say that atomic contains a T, which we're working on for the next mailing. Then we can fix this more easily example (perhaps even in the same paper).

[jensmaurer]

Editorial meeting: Yes, need SG1 input.