Odd "endianness" [was Re: RE: Base 64 posts to the list]

[Charles Anthony]

On Mon, Dec 5, 2016 at 11:14 AM, Dave Wade <dave.g4ugm at gmail.com> wrote:

> > -----Original Message-----
> > From: cctalk [mailto:cctalk-bounces at classiccmp.org] On Behalf Of David
> > Bridgham
> > Sent: 05 December 2016 18:37
> > To: General Discussion: On-Topic and Off-Topic Posts
> > <cctalk at classiccmp.org>
> > Subject: Re: Odd "endianness" [was Re: RE: Base 64 posts to the list]
> >
> > On 12/05/2016 12:17 PM, Chuck Guzis wrote:
> >
> > > Or how about architectures not using a word length that's an integral
> > > number of bytes?
> >
> > You mean like any 36-bit machine?
>
> Honeywell L66 & DPS8 used to have 36 bit words which originally contained 6
> x 6-bit characters.
> When they extended the machines to work with ASCII they put 4 x 9-bit
> characters which I seem to
> remember they called 9-bit bytes..
>
> Yes; the Extended Instruction set handles 4*9bit, 6*6bit, 8*4bit (with the
4 padding bits scattered through the word).

>From memory:

PDP-10: 36 bit word, 5*7bit characters.

PDP-15: 18 bit word, but it was so long ago, I don't remember....

CDC 6000: 60 bit word, 10 six bit characters.

PDP11 "middle endian" see "NUXI problem:
http://www.catb.org/jargon/html/M/middle-endian.html

PDP11 RADIX-50 3 characters packed into a 16 bit word; each character in a
0:39 set.

Back to Der Mouse question re: non-symmetrical mapping....

hton and ntoh are not meant has generalized data conversion; they are
intended as network data packet field conversion; the domain of ntohl is a
32bit unsigned integer; the range is a host object larger enough to contain
all possible values.

For hosts that are base 2 and have word sizes that divide 32bits evenly,
 the functions would typically be identity  or bit rearrangement, and the
htonl and ntohl functions would be symmetric -- I suspect that a good
mathematician could 'prove' that cycle length is always 2 given the
constraints.

Cases like 36 bits words mean that htonl is "lossy"; it throws away bits
and and ntohl pads the result with 0s -- they are not symmetrical, thus the
answer to the order-2 cycle question is 'not applicable'

-- Charles