Steve Johnson scj at yaccman.com
Tue Sep 19 02:46:29 AEST 2017

When we decided to port Unix to 32-bits, there were a number of
candidates we looked at, including the IBM-360, the DEC-20, and the
Interdata.   The Vax was barely there at the time, and we also felt
a bit of pressure to use a non-Dec machine because some manufacturers
were muttering about a too-close association of ATT & Dec that was
incompatible with our status as a regulated monopoly.  It was also
clear that being able to buy compatible Unix machines from several
different manufacturers would ultimately drive the prices down and be
good for ATT.

Dennis hated the channel I/O on the 360, and I hated the little
pinhole glimpses of memory that could only be seen by loading a
register--that really clashed with the "one flat memory" model in C. 
For the DEC-20, the word size was indeed a problem.  We had built a C
compiler for the 36-bit Honeywell machine, and it was quite happy to
have 9-bit bytes (in fact, their OS supported this).  But at the
time, tape drives were the major way large data got into and out of a
machine, and they all liked 8-bit bytes.  And there were issues in
both C and Unix if the byte size didn't evenly divide the word size
(think malloc, unions, sizeof, etc.).  So the 20 was out.

The Interdata was a good compromise.  The machine had been used in a
lot of flight simulators, etc., so it had a track record.  It was
similar to an IBM-360, but had an address mode for addressing memory
directly with no base registers.  And we bought the machine with
writable microcode, which we had some ideas of playing with.

As it turned out, the biggest problems in porting were taking
little-endian code and data and putting it on a big-endian machine,
dealing with the consequences of having more registers, and the effect
of this on process switching and debugging (stack traces, etc.).  The
pre-V7 system calls liked to return -1 when there was an error -- V7
wanted to see nulls.  This turned out to be a critical problem -- if
an old-style call with -1 got returned and taken to be a pointer, the
Interdata generated a double fault--trying to read a word with an
odd-byte address, and memory out of bounds.  Sadly, these faults came
in several subtics apart, and the result was that the address of the
fault got trashed, leaving the machine to resume at location 0 with no
hint about how it got there.

For this and other reasons, Ken and Dennis invented the header file
concept -- previously, the structs that described data structures like
inodes were printed in the manual, and copied out by hand when they
were used.  By this point, Lint was a going concern, and I hacked a
version of Lint that insisted that when a structure was declared more
than once, the declarations must come from the same physical
location.  This was a big help in rooting out hidden obsolete
structure definitions and forcing the use of header files.

The port ended up being fairly successful.  The connection with
Interdata less so.  At one point there was a meeting attended by me,
Dennis, Ken, and several managers.  We went to Interdata and met in a
building that used to be a bank -- the meeting room was in the old
vault.  We told Interdata what we had done, and  gave them some
information about the growth of Interest in Unix.  We then said that
the machine needed some fixes before we could deploy a general-purpose
system on it -- the -1 botch was the biggie, because we could not have
a broken user program turn the machine into a stone.  Interdata
listened politely.  Several days later they said 'not interested".

One thing I noticed over the years is that there are very few people
who, having done one Unix port, were willing to do another.  I
certainly wasn't interested.  After the first time, there is little
to learn, and the tedium of debugging a compiler and OS on a bare
machine, when the documentation of the machine was hastily written and
often incomplete, was frustrating almost beyond describing.  So it
was another group in Bell Labs that jumped on the Vax and made an
effective and lasting 32-bit port.


----- Original Message -----
From: "Arthur Krewat" <krewat at kilonet.net>
To:<tuhs at minnie.tuhs.org>
Sent:Mon, 18 Sep 2017 11:30:19 -0400
Subject:Re: [TUHS] PDP-10 UNIX?

 On 9/17/2017 10:34 PM, Johnny Billquist wrote:
 > On 2017-09-17 18:33, Arthur Krewat <krewat at kilonet.net> wrote:
 >> Was there ever a UNIX or even the thought of porting one to a
 > Definitely a thought. An attempt was started on NetBSD for the
 > and it sortof got halfway of getting into single-user, but I'm not 
 > sure if the person who worked on it just got distracted, or if he
 > problems that were really hard to solve. I certainly know the
 > and can find out more if people really are interested.
 I for one would be VERY interested. It would be an interesting
 to try it out on a SIMH emulator.
 >> 8-bit bytes would have been a wasteful exercise, but you never
 >> (losing 4 bits of every 36-bit word)
 > Uh... Why 8 bit bytes? That way lies madness. There exists a really

 > good C compiler for TOPS-20 - KCC. It uses 9 bits per byte. Works
 > a charm, except when some people write portable code that is not so

 > portable. ;-)
 > KCC was written by KLH, unless I remember wrong. Same guy who also 
 > wrote the KLH-10 emulator.
 9 bits would work - and still, waste 4 bits of the 36-bit word ...
 code carried over from elsewhere would never use the 9th bit unless
 know of stuff that does?

 And yup, I remember KLH, not personally, but as the KLH-10 author :)

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://minnie.tuhs.org/pipermail/tuhs/attachments/20170918/fb60ed1f/attachment-0001.html>

More information about the TUHS mailing list