1) use reinterpret_cast<> as I proposed, but documenting that it will produce a viable hash function only if the underlying implementation satisfy certain requirements (each specific implementation /should/ document the behaviour of reinterpret_cast anyway, so the user may able to check)
2) provide the hash function (with reintepret_cast<>) only for those platforms that we are able to check in advance that will satisfy those requirements and make any use of such function produce a static assert for all other platforms
3) do not provide a hash function for pointers at all
I would vote for 1, but 2 and 3 also make sense. Is there a fourth choice?
I would also go with (1), and probably a cast to ptrdiff_t is sufficient (rather than int_max_t). There is a forth option: treat the pointer as a "bag of bytes" and hash those bytes, that's completely portable, and no reinterpret_cast is required (it's probably overkill as well) John.