Endian order

[dacut at neolinear.com]

> Adjustable byte order is an optimization, same reason dbus uses a binary
> protocol. For a heavily-used library in low-latency local IPC, avoiding
> the byteswaps seems worthwhile, even if it involves more code in the
> implementation.

Ah, actually, I remember one other reason I chose network-byte order: it's
the native ordering on Sun SPARC boxes.  Most of these are Blade 100's; any
processing I could push off onto the Intels was for the better. :-)

(EDA is slowly migrating away from Solaris/Sun to Linux/Intel, but it'll
take a couple years before x86 is the dominant platform...)

> Of course you'd have to run profiles to prove it one way or the other.

Ugh.  That was one of the most unpleasant tasks I've undertaken -- trying to
figure out "which implementation runs faster" across two machines on a
corporate Ethernet.  It seemed every datapoint was an outlyer.