Xref: utzoo rec.humor:18616 rec.humor.d:1569 comp.misc:4966
Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!csd4.milw.wisc.edu!bionet!agate!saturn!ucscc.UCSC.EDU!haynes
From: haynes@ucscc.UCSC.EDU (Jim Haynes)
Newsgroups: rec.humor,rec.humor.d,comp.misc
Subject: Re: Looking for Computer Folklore
Message-ID: <6255@saturn.ucsc.edu>
Date: 8 Feb 89 04:56:51 GMT
References: <7143@pyr.gatech.EDU> <532@geovision.UUCP> <4575@tekgvs.GVS.TEK.COM> <799@n8emr.UUCP>
Sender: usenet@saturn.ucsc.edu
Reply-To: haynes@ucscc.UCSC.EDU (Jim Haynes)
Organization: California State Home for the Weird
Lines: 44

Time for some DEC stories.

A lot of the PDP11s had optional floating point hardware.  The assembler
program had an option flag so that it could either assemble floating
point instructions or assemble calls to a library of software routines
simulating the floating point hardware.  Well, once upon a time they
released a set of diagnostics for the floating point hardware that had
been accidentally assembled using the software simulation flag, so they
didn't exercise the hardware at all...

We had a system start crashing frequently, trashing the system disk in
the process.  After a lot of very painful examination by hand of disk
contents we concluded that it was occasionally missing a word
transfer, or doing an extra one, depending on the direction of transfer.
This we traced to a bad Unibus receiver chip that was allowing a glitch
to get through as a legitimate pulse.  The annoying thing was that the
disk diagnostic program never detected the problem, because it wrote
and read back a block of identical words, and didn't clear the read
buffer between reads.  When every word looks like every other you don't
notice the missing or extra ones.

Another G.E. story: nothing wrong here, just an interesting consideration.
In MULTICS the customers wanted the calendar time kept in something like
microseconds since midnight, January 1, 1901.  We had a whole rack of
equipment just for this clock, which used something like a 72-bit
register.  Some of the most significant bits of the clock wouldn't
change for years after it was built; so we had to have logic that
would copy the value to another register, flip all the bits, verify
that they flipped, then flip them all back and verify that they all
flipped back.

I had occasion to look at the date routines in an early version of the
operating system for the Burroughs 6500.  Not only did it take care of
leap years every four years; it correctly handled the exceptions for
centuries and millenia.  If you're going to write software that is
correct, you might as well write it to be correct for the next few
thousand years.  I guess you don't worry about the extra computation if
it's only executed once a day.
haynes@ucscc.ucsc.edu
haynes@ucscc.bitnet
..ucbvax!ucscc!haynes

"Any clod can have the facts, but having opinions is an Art."
        Charles McCabe, San Francisco Chronicle