Re: interesting

From: Clifford Heath (no_at_spam.please.net)
Date: 03/16/05 

Date: Wed, 16 Mar 2005 17:58:24 +1100

Terry Given wrote:
> CS is a lot easier to outsource than EE,

Disagree that CS is easy. It's too hard to adequately specify
software compared to hardware. The only software which can be
safely outsourced should be produced by a compiler operating
directly on the specification. I should know - I've written
more than one such compiler and have more in mind :-).

> ..as the number of tasks an instrument performs increases,
> the required complexity increases much, much faster.

True. And as complexity increases, the "software entropy" curve
trends upwards. By software entropy, I mean the ratio of the
complexity of the software to the complexity of the problem.
The higher this ratio, the faster it grows.

> Absolutely. What makes matters worse is that the techniques for writing
> good OS' have been known for DECADES.

Some of them, but not enough.

> Dip*** workers screw up almost everything they touch.

Seen a few of those, but I only remember in their backsides ;-).

> I think a lot of sw problems are because managers often understand
> little or nothing about development of decent software,

That's not mainly their fault, it's because we've been pretty bad at
describing and quantifying quality.

> This is where a background in software engineering is very useful - eg
> data structures can (and should) make an early entrance into the design
> of a piece of sw, with little or no consideration of how they might be
> implemented.

Data structures are about implementation. The discipline you
mean is called "data modelling", and is one of the main things
I harp on about and teach people. It's relevant that UML seems
to have been created to make it impossible to do properly. I
believe in object-oriented design, but I resile from the notion
that proper object modelling means you *aren't allowed* to do
data modelling. This ignorant notion comes from acknowledged
leaders in the field, who maintain that if you fully explore
the behaviour and relationships between objects you can keep
their internal states hidden, even from the designer. Nothing
could be further from the truth, and the only reason such
projects ever succeed is the designer has a good perception of
the unspoken states and their required representations.

Clifford Heath.

Quantcast