Re: Top 10 for old farts

[davidf(at)dfanning.com (David Fanning)]

Since I enjoy (as Brian Jackel puts it) a "pointless, but
interesting discussion" almost as much as I do writing IDL
object programs, let me just weigh in with one more response.

Pete Riley (uk2@linuxfan.com) writes:

> Why not replace all the direct graphics calls with object 
> equivalents? You should be able to say:
> 
> PLOT, findgen(10)
> 
> and you get an object window (maybe plus the print, modify,etc...options
> that are in insight). There should have been a transparent migration to
> object graphics. 

This was exactly RSI's original intention. But, alas, things
don't always go the way we plan. And with object graphics the 
vagaries of programming projects were even more pronounced
than usual. It was almost as if Coyote himself were put in
charge of it.

Let me recount the story as I remember it. (Which is to
say, this may not be how it *really* happened. But it is
the way I tell the story.)

Some four or five years ago now RSI realized that their
graphic system was out of date. It was wonderful for the
early 1980's, but these were the 90's and the millennium
was fast approaching. And in particular, their 2.5D way
of rendering surfaces (the Z axis must always be vertical)
was hopelessly out of date. Three orthogonal axes should
be completely rotatable. The decision was made to completely
overhaul the entire graphic system and bring it up to date.

One of the best 3D graphic guys in the country (I forget
his name) was hired to design the system and the RSI
programmers set to work. A year was set aside for the
effort, during which time there would be no work done
on bug fixes etc., as everyone was tied up in building
this new thing. Customers would get jumpy about paying
maintenance dollars during this time, but the new system
would be so much better than the old that this problem
was judged worth the risk.

To make a long and painful story short, it took a LOT
longer than one year to do the work. In fact, it was
nearly two years by the time IDL 5.0 left the barn.
Customers were screaming, maintenance revenue was
dwindling, new license sales were off as customers
were waiting for the big new release that was right
around the corner. And expenses were up sharply as
more programmers were hired to keep up with the
larger and larger work load. I think everyone in the
building was putting in 60+ hour weeks. In short, it 
was one of those really tense, trying times that 
all businesses go though occasionally.

But about 6 months before the actual release the graphics
system was built sufficiently (most of the effort up to
then had gone into coding the graphics primitives that 
could be assembled into line plots, surface plots, etc.)
that simple "high level" graphics commands could be
tested. The idea had always been (I think because this
is what the graphics guy told us) that 2D graphics would
"fall out of" the 3D graphic system. In other words,
2D graphics were just a specific subset of the 3D system.

But that turned out to be a bit of an oversimplification.
Because when the first line plot was tested in the new
graphics system it rendered about 10 times slower than
the equivalent direct graphics command. In fact, *all* of
the 2D commands were about 10 times slower than the
equivalent direct graphics commands. The problem, as
it turned out, is inherent in 3D systems, and cannot
be easily solved even today, I think.  It is the
reason you have never been able to make a decent line
plot in a fabulous 3D graphics system like AVS.

Oh, oh. We were going to go to angry customers and tell
them "OK, you paid us a lot of maintenance money for
two years and didn't receive anything for it, but here
is your great new system. And by the way, it is 10 times
slower than the old one." It was going to be a hard sell,
to say the least. :-)

So the decision was made (the correct one, I think) to
keep the old direct graphics system, but to add the new
object graphics system to IDL. Customers could use the
one that made the most sense to them. It was expected
that most line plots and image displays would be done
in the old direct graphics system (since it was a lot
faster than the object graphics system), but that 
visualizations that required a robust 3D capability would
use the object graphics system.

That is pretty much where we are today.

Where RSI has fallen short, I think, is in not making
more high-level object graphics commands. It is still
pretty much a system that has to be built from scratch.
(I know there is IDL Insight. But have you used it!?
I find it infinitely frustrating and completely non-intuitive.
Plus, aesthetically it is awful. I'm embarrassed by it 
whenever I have to show it to someone.) And no one else
has stepped forward really to make the kinds of tools
that people need. (Partly, I should tell you, because
no one seems to want to pay you for making the effort. :-)

My sense of what is happening in object graphics is that
RSI is moving on to other things (I don't see much in
the way of updates on the list of features for IDL 5.4.)
Perhaps they are locked into Insight for political
reasons, I don't know. Or perhaps there are more people
using it than I realize. (Lord knows I couldn't abide
a Top 10 Things I Like About Insight thread.)

But, in any case, objects (which had to be invented
to create the object graphics class library) are NOT
object graphics. Too many people fail to make the
distinction. When I talk about object programming
I am very rarely talking about using the object graphics
class library. I'm taking about using a new way of 
constructing programs that has the potential to make
your programs much more useful to you.

This ability to create your own objects is the
major story, I think, in the IDL 5.0 release. Not
the object graphics systems. I don't know why it
has fallen on me to be the spokesman for the
Crusade to Get the Object Story Out, but it appears
to be the case. But then Coyote has always liked his 
little jokes. :-)

Cheers,

David

-- 
David Fanning, Ph.D.
Fanning Software Consulting
Phone: 970-221-0438 E-Mail: davidf@dfanning.com
Coyote's Guide to IDL Programming: http://www.dfanning.com/
Toll-Free IDL Book Orders: 1-888-461-0155