lfpinit() signed or unsigned?

[Eric S. Raymond]

Hal Murray <hmurray at megapathdsl.net>:
> 
> Stromeko at nexgo.de said:
> > Does anybody know of an implementation that does _not_ preserve the
> > representation when converting from signed to unsigned integer of the same
> > width?  The assumptions already made are: Corresponding signed and unsigned
> > representation have the same number of bits, signed uses 2's complement
> > representation, there are no extraordinary values and no padding bits in the
> > representation. 
> 
> I think all modern machines are using 2s complement arithmetic.
> 
> 50 years ago, a few systems used signed-magnitude and 1s complement.
> 
> That "implementation dependent" sort of weasel wording is probably
> left over to cover that case.  It might actually get interesting if
> somebody ported a modern c compiler to work on one of those older
> systems or an emulation of one.

To my utter, gob-smacked astonishment, it turns out this is still a
live issue. Though, thankfully, not for us.

I thought I could just mumble something about the CDC 6600 and be
done, but if you google "one's-complement machines" you'll find that
Unisys still ships a series of machines with the brand "Clear-Path
Dorado" (latest variant introduced 2015) that are emulations of their
old 1100-series mainframes running over Intel Xeon hardware - and
these have one's-complement arithmetic.

This isn't a practical port blocker for us, as NTP will never run over
the batch OS on these things - it's about as POSIX-compatible as the
Bhagavad-Gita. It's just weird and interesting that one's complement
machines survive in any form at all.

And a bit personal for me.  My father was a programmer at Univac in the
1950s and early '60s.  He was proud of his work.  My very first interaction
with a computer ever was getting to play a very primitive game on the
oscilloscope-based video console of a Univac 1108.  This was in 1968.  I
was 11 years old, and my game machine cost $8M and took up the entire
ground floor of an office building in Rome.

Other than the 1100, the one's-complement machines Wikipedia mentions
(LINC, PDP-1, and CDC6600) are indeed all long dead. There was a
one's-complement "CDC Cyber 18" as late as the 1980s, but again
this was never going to implement POSIX.

There's a comp.lang.std.c article from 1993 that argues in some technical
detail that that a C compiler is not practical on one's-complement hardware
because too many C idioms have 2's-complement assumptions baked in:

https://groups.google.com/forum/#!topic/comp.std.c/GlZeSxBnCW0

I think I believe this.
-- 
		<a href="http://www.catb.org/~esr/">Eric S. Raymond</a>

Please consider contributing to my Patreon page at https://www.patreon.com/esr
so I can keep the invisible wheels of the Internet turning. Give generously -
the civilization you save might be your own.