Re: [GD-General] Architecture Design

[Brooks B. <br...@on...>]

>You mean, like Michelangelo?  :)  (Isn't he the one who described his
>process that way?)

In the Smithsonian there is a "cartoon" done by Michelangelo (I 
think...maybe it was Rafael?).  Cartoons are design sketches done by the 
artist prior to beginning an actual painting.  The sketch is worked over 
heavily - things are erased and moved around until the artist is happy 
with the final design.  The cartoon is very messy.

At that point the artist punches holes all along the outlines of the 
cartoon.  They then attach the cartoon to the final media (wall for a 
mural, canvas for a painting) and throw charcoal dust on it.   The 
charcoal goes through the holes and leaves a connect-the-dot framework 
that the artist then paints on.

Sooooo....  there is some level of framework design in paintings of the 
masters.  I was really suprised - I just thought they were so good they 
just winged it.  (Also, since the cartoons were destroyed usually, they 
are incredibly valuable today).

>Back on topic, the bad pitfall I've seen in detailed top-down engine
>designs, is when you break things down into sensible pieces with sensible
>interfaces, and some of the boxes implicitly contain things like "robust
>rigid and soft body physics solver", "automatic all-purpose 3rd person
>cinematic camera", "human-like real-time learning neural net AI".
>

I call this the fractal code coastline problem.   You can keep breaking 
things down. But you can't see the total detail until you have broken it 
down to the atomic level of code.  

The worst thing about design work is that it is tempting (for management) 
to look at all those boxes and assign times to them to arrive at a final 
schedule.   But you can't tell how long the boxes take unless you break 
them all the way down to code again.  So you miss the fractal multiplier 
that should be applied. 

>think about it this way, the russians were 20 years ahead of the americans
>at designing rocket boosters, but their designs took longer to implement, so
>whilst the americans went to the moon, the russians didnt

One of the more interesting aspects of the American effort is that Von 
Braun and company skipped the detailed QA cycle.  They were so confident 
in their design, and time was so short, that they simply test fired the 
final completed Saturn rocket.  It worked. This saved a tremendous amount 
of time.  Had the rocket failed it would have been a huge setback.  

The Russian designs did not work out of the gate.  So in a way the speed 
of the US program is probably more a result of careful conservative 
design than of aggresive forward thinking design.