[atomics.types.operations] Possibly unintended example

[frederick-vs-ja]

The example in [atomics.types.operations] Note 8 currently reads:

union pony {
  double celestia = 0.;
  short luna;       // padded
};
atomic<pony> princesses = {};

bool party(pony desired) {
  pony expected;
  return princesses.compare_exchange_strong(expected, desired);
}

In common implementations, pony have the same size as double and a default- or value-initialized pony doesn't have any bit with indeterminate value. So it's questionable to me whether this example serves its purpose. Would it make more sense to initialize the short member instead?

[eisenwave]

Would it make more sense to initialize the short member instead?

I think the point is that princesses could have luna active at some point, so even if expected is zeroed, the compare_exchange could fail endlessly. It's debatable whether this is more surprising or less surprising than if expected had luna active.

Or maybe, I misunderstand the example and even if celestia is always active, it can fail, but I don't think I do.

Unless that's the case, the example seems more or less okay to me, but I wouldn't oppose being a bit more explicit about why the comparison can always fail in the note, or using luna as an active member.