The transformations are placed on a stack, and then applied from the top
down. So the most recent transform is applied first.
This highlights something I've meant to gripe about for some time.
The VRML spec as presented assumes knowledge about Inventor. The
transform example is one case. Another is how the camera is aligned.
It seems unacceptable for me to expect people to know about Inventor
in order to use VRML. It is counterproductive to the aim of establishing
VRML as the standard for web 3D.
Still on the camera example, even now I can't figure out how the f#$% it
works. I've wasted so many hours on it that I've been forced to give up.
That results in a modeller that produces strangly behaved scenes, but I
cannot offer more.
To make matters worse, the open spec means that every browser (and I
literally mean every) does something differently. We had Guy Steele
here last week talking about Java, and the current process of tightening
the Java specs before release. He emphasised that everything must be bolted
down firmly, because any possibility for variation will lsad to
incompatability among browsers, and that is enough to kill a standard.
I think that's a vote for a library of test scenes, which is finally coming
into fruition. It is also a vote for a tight standard. Before the flame wars
begin, I do realise that one of the aims of VRML is to put it out there
and see how it evolves. My point is that with insufficient agreement it
cannot evolve.
Steve.