[TUHS] Portability (was: BSDi Imaging)

Steve Johnson scj at yaccman.com
Mon Feb 27 11:08:39 AEST 2017


I couldn't disagree more.  Here is some background.

I didn't have much experience with COBOL or Algol -- Bell Labs was a
FORTRAN shop.  And we had a strong need to run FORTRAN programs on
many different kinds of computers.  So the portability of FORTRAN
programs was very important.  Barbara Ryder built a program (in
FORTRAN!) called the PFORT verifier that verified that a FORTRAN
program could be run on any of the six dominant machines of the day
(this was an inspiration for Lint, by the way).  It was incredible
what differences she found in this "portable" language when different
implementations encountered the programs.  My favorite was a bug in
one compiler that would cause a program that began with 50 comment
cards to abort.  The FORTRAN program produced a title line on the
listing (remember those?) that was printed out, and it used the first
function name to get the title.  50 lines => new page, no title =>
blam.   The subtle bugs caused by copy-in/copy-out argument passing
were especially icky.

So portability was much in our minds in the late sixties and early
seventies.  As Unix began to catch on, there was a desire to move
some utilities to other systems.  Neither B nor the original C was
really up to the job.  Remember that machines of the day had
different character sets, different floating point implementations,
and different byte and word sizes, word address vs. byte addressed,
and with different word orders.  It was a zoo.  Moving programs in
any language was difficult, not least because the operating systems
were so different (remember JCL?).

Late in 1974, as I recall, Dennis mused "You know, I think it would be
easier to move Unix to a new machine than to change a large
application to run on another operating system."  I thought about
this for a few minutes, and then offered to write a portable C
compiler.  I had already moved B and/or C from the PDP-11 to the
Honeywell (36-bit words, 9 bit bytes, word addressed) and the IBM/360
(32-bits, byte addressed, but bizarre difficulties accessing memory in
a reasonable fashion), so I had some idea what I was getting into.  
We also persuaded the company to buy us a 32-bit minicomputer (the
Interdata 4/32) with the goal of porting C and Unix to it.

With so many different kinds of opcodes and data formats, Dennis made
what I think was an inspired solution.   + meant "add" on whatever
computer it was running on.   int was the natural word size.  etc.
This had three very good properties:
1) it kept the language simple.   In particular, it kept us from
having to specify bit widths for everything, which would then need to
be changed when the program was ported to avoid inefficiency.
2) It kept the compiler simple.  This was essential on the small
PDP-11.
3) It turned out that with the addition of Lint, many problems that
could result from porting could be found, without preventing the
program from running well on the host system.

So C was indisputably intended to be portable, at least in that
sense.  And in practice it was highly portable while sacrificing
little in performance on different systems (unlike some other
languages).

There were drawbacks -- it wasn't perfect.  Cross compiling from one
machine to a much different one still had its challenges.  Adopting
the target machine's byte order made networking code harder to
write.  And machines that considered character constants to be signed
produced a lot of irritation (and a lot of Lint messages).  But it
wasn't from lack of trying...

Steve

----- Original Message -----
From: "Greg 'groggy' Lehey" <grog at lemis.com>
To:"Noel Chiappa" <jnc at mercury.lcs.mit.edu>
Cc:<tuhs at minnie.tuhs.org>
Sent:Mon, 27 Feb 2017 07:49:00 +1100
Subject:[TUHS] Portability (was: BSDi Imaging)

 On Sunday, 26 February 2017 at 7:50:50 -0500, Noel Chiappa wrote:
 >
 > There were in theory portable languages beforehand (e.g. PL/1), but
 > I think it probably over-specified things - e.g. it would be
 > impossible to port Multics to another architecture without almost
 > completely re-writing it from scratch, the code is shot through
with
 > "fixed bin(18)"'s every other line...

 That may be coloured by your perspective. C was never designed to be
 portable, while much older languages like Algol and Cobol were. There
 were quite different reasons for C's success.

 To quote "The Programmer's ABC'sâ from Datamation, April 1976:

 C is for Cobol
 What a pity
 It was designed
 By a committee

 Greg
 --
 Sent from my desktop computer.
 Finger grog at lemis.com for PGP public key.
 See complete headers for address and phone numbers.
 This message is digitally signed. If your Microsoft mail program
 reports problems, please read http://lemis.com/broken-MUA


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://minnie.tuhs.org/pipermail/tuhs/attachments/20170226/a890d04c/attachment.html>


More information about the TUHS mailing list