Computer Mathematics was Re: [LUG] accountancy experience required

[Simon Waters <Simon@xxxxxxxxxxxxxxxxxxxx>]

Adrian Midgley wrote:
> On Sunday 04 August 2002 11:11, you wrote:
>
> > this would mean that it would be very difficult for accountants to
> > ignore/reject sql-ledger - which is web enabled and open source.
>
> A typical question is: "is it validated".  Which I always find hard to answer
> - since "I've checked it carefully" doesn't seem to be the approved answer<g>

Ask them what they mean by validated ;)

I wouldn't totally trust anything that comes out of a computer,
although they are much more reliable than people, I've seen too
many CPU's with floating point errors, and the like.

For example, one of the alleged ways of detecting the Pentium
floating point bug, was actually revealing a bug in Microsoft's
calculator program.....

Now I have to say at integer maths, computers are a lot better,
and most of accountancy is integer mathematics, but I doubt
everybody codes accountancy using integers, programmers aren't
that uniformly well trained. Everyone here knows how their
spreadsheet represents currency internally, and are sure that
every field with currency in is in the right format, don't they
?

The silliest maths story that happened to me, was when I worked
on a program calculating aircraft height from reported pressure.
When a pressure of zero was reported, one computer crashed, the
other program issued a warning and continued with height of
+INFINITY (which is of course totally correct, and totally
useless), the difference was purely architecture IBM 390 v IEEE,
the software was the same Fortran subroutine.

So even 100% code coverage won't cover you if you don't specify
the environment in which it is run precisely enough.

Another revealing experiment was the climate forecasting on the
Cray. They would expect runs of the computer model to bit
compare, as proof the software was not changed unexpectedly by
certain changes to the IT.

This level of attention to detail revealed;

That 1/X worked out long hand differed from the Inverse of X
using the vector processor inverse hardware on that Cray.

Cray correcting some basic trig functions due to bugs in the
implementation of the maths libraries. They backed out these
corrections till they completed that series of experiments,
reproducibility is sometimes more important that accuracy.

Once you've got down and dirty with the innards, you'll never
look at computers as infallible ever.

Simon

--
The Mailing List for the Devon & Cornwall LUG
Mail majordomo@xxxxxxxxxxxx with "unsubscribe list" in the
message body to unsubscribe.