Doing numerical computing requires special skills that are rather
distinct from what they used to call systems programming. For example,
programming an inverse chi square distribution function that works
accurately and efficiently is quite different from writing a function
that deals only with string and integer quantities. The required
skills are taught in computer science departments, but probably the
people who managed Excel didn't know anything about them and didn't
care. It's really a sign of sloppy management. The litany that
Linonut cites is pretty damning.
I know a guy who is on a crusade regarding floating point processors.
The standards developed for these things in the 80's took a lot of
thought, and in some cases were painful for the engineers to implement.
Now these standards are being abandoned by manufacturers (IBM was
mentioned in a talk I heard), because they're too much trouble to
implement and because for gaming applications it doesn't matter if a
floating point operation makes an occasional error. But these same
chips are also being advertised as being useful for number crunching
(designing bridges, space shuttles, etc).
|
|
