🚀 go-pugleaf

In article <JBW.91Jul19164859@maverick.uswest.com> jbw@maverick.uswest.com (Joe Wells) writes:
   Wrong!  It is obvious, when "one looks at the larger picture".  Your
   method would require every computer to know what byte order every other
   computer used, which is impractical.  The only practical solution is to
   use a canonical byte order, and it doesn't matter which one you pick.

You have the choice of describing byte order along with the data,
or picking a canonical byte order.

The first method has the advantage that data transfer between similar
machines requires no byte swapping and, thus, is faster. A library of
vision utilities that I wrote uses this method (should I patent it? :-).
However, it makes the reading routines more complex, since they must
be able to deal with both byte orders.

The second method saves the storage overhead of indicating the byte
order, and it means that any I/O library only needs to have code
for the one standard byte order (in terms of programming effort,
there is no savings, since you need to write code for reading and
writing for machines of either byte order anyway). The disadvantage
is that for transfers between machines of identical byte order
(the usual case!), you often incur unnecessary overhead for byte
swapping.

The first method is clearly preferable if you want highest
performance. The second method is a little simpler in practice.

						Thomas.