+1 to this proposal. AFAIK it solves everything we discussed recently, does this to a much greater level of detail and is backed with the implementation already, so well done.
memcpy
works at the byte level (i.e. it can't copy 7 bits of a byte without copying the 8th (memory I/O is done at a byte granularity)) and thus memcpy doesn't even talk about bits in its man page. However I agree that "BitwiseCopyable" is a good name to use here and not any worse than a potential "BytewiseCopyable" alternative.
I guess Double could follow the suit.
What about the tuples? Will there be some automatic conformance that says: the tuple of "BitwiseCopyable" elements is "BitwiseCopyable" itself? Or will it be possible to manually conform a tuple class to "BitwiseCopyable" (provided its elements are "BitwiseCopyable")?
On a point of clarification, would the "BitwiseCopyable" applied to a struct mean what "POD" currently means? In other words will it follow (1) or (2) below?
On that front, could BitwiseCopyable be applied to a reference type? (obviously the one that doesn't contain reference fields in its payload). I guess – "no" and it is worth to mention it in the proposal (unless it is already mentioned and I just missed it).