Byte Order Swapping Function

[Timon Gehr]

Jonathan M Davis wrote:
> On Sunday 17 July 2011 09:40:06 Alix Pexton wrote:
>> On 17/07/2011 07:42, Jonathan M Davis wrote:
>> > On Saturday 16 July 2011 23:31:09 Andrew Wiley wrote:
>> [snip]
>>
>> >> Take a look at http://www.dmh2000.com/cpp/dswap.shtml . It made the
>> >> odd
>> >> behavior make a lot more sense to me.
>> >
>> > Okay. Good to know. In other words, we can't have swapEndian work with
>> > floating point values like that. It can generate a byte array from them
>> > (or maybe an integral value of the same size), but it can't generate
>> > another floating point value. Bleh. Okay. Floating point values with
>> > have to use an entirely different overload then. Thanks for the info.
>> >
>> > - Jonathan M Davis
>>
>> I find myself wanting to address this issue by using types, as the FP
>> example demonstrates that when the endieness of a value changes, its
>> type should too.
>>
>> using double as an example, I'd like to be able to write code something
>> like this.
>>
>> > auto v0 = 0.1;
>> > auto v1 = nbo!double(v0); // nbo = network byte order, your names may
>> > vary auto v2 = to!double(v1);
>> > assert(v0 == v2);
>>
>> I would, of course, expect aliases for those occasions when one wants to
>> state endieness explicitly (I just can't think of any >< ), and for
>> non-conversions (when nbo == hbo) to be no-ops.
>>
>> Or am I missing something?
>>
>> A...
>
> I suppose that you _could_ do something like that, though I'm not sure that
> there's much point. You can swap integers with impunity. Having a separate
> type would actually be a problem, because then you'd have to translate to an
> integer again anyway to do anything with it. In the case of a floating point
> value, you essentially have to convert to either an integral value or an array
> of ubytes. You could wrap that in a struct or somesuch, but I'm not sure how
> much point there is in that. Just like with integers, you'd have to convert it
> over to a prmitive type before doing anything with it like sending it via the
> network.
>
> So, sure you _could_ make such types, but I don't think that there's much
> point to it. And if someone wants to do that in their own code, it wouldn't be
> all that hard. The hard part is the actual byte swapping.
>
> - Jonathan M Davis

Imposing meaning to bytes is the only point for data types to exist whatsoever.
How do you distinguish between an integer and a network byte order integer at
function boundaries?

Or, in other words:
Endianness-swapped integers and integers are two very different things and ought
not to be interchangeable.
Sure, you have to convert them to ubyte[] or integer if you want to send them over
a network, but that is true for every datatype.
An integer is a number stored in the native endianness of your system. You
shouldn't be able to use an endianness-swapped int as an int in your code. An
endianness-swapped integer is an entirely different beast.


I think Alix' suggestion is the way to go.

(Of course, for integers you would also want to provide the byte swap functions,
if somebody does not actually care about endianness and networking/serialization,
but just wants to swap the byte order.)


Cheers,
-Timon