|
A natural expansion of the basic Node definition is the addition of
a third dimension.
There are many ways to achieve an illusion of three dimensions within
a two dimensional surface, namely, the computer screen.
The movement of the node is easily changed. We need only duplicate
a set of transforming equations and change the variable names.
The method outlined here is not a true three dimensional perspective,
but it's convincing, easy to implement, and doesn't require too many
additional resources.
First, we'll take care of the node's movement by adding variables
z and vz to their
respective transforming equations.
|
|
|
Accelerate the Node through three dimensions
with friction
|
|
|
// calculate acceleration relative to destination
var fx=(this.dx-this._x)/50;
var fy=(this.dy-this._y)/50;
var fz=(this.dz-this.z)/50;
// accelerate towards destination
this.vx+=fx;
this.vy+=fy;
this.vz+=fz;
// apply inertia
this._x+=this.vx;
this._y+=this.vy;
this.z+=this.vz;
// apply friction
this.vx*=.7;
this.vy*=.7;
this.vz*=.7;
|
|
|
Now to make the dimensionality of the node convincing, we need to
do two things: stack the nodes within the layers and scale the nodes
relative to their depth.
Stacking the nodes can be brutal. In the Flash environment, each layer
is allowed only one object. Objects in higher number layers are shown
on top. Placing an object on a layer destroys anything that might
have been there before. Instead, we would like to have nodes to be
able to exist simultaneously at the same depth without clobbering
each other.
The solution is to track which layers have objects, and space out
the mapping of depths to layers by a factor of about 10. That is,
for each integer depth, there are 10 layers available for object placement.
This system will work only if we constrain values of z (the depth)
to a reasonable, positive range (say 0 to 100).
We'll keep an array, filled initially with all zeros, to indicate
which layers have objects placed within them.
|
|
|
The nodeList array
keeps track of what layers contain a Node |
|
|
maxNodes=1000;
nodeList=new Array();
// set up depth ordering index
depthList = new Array(maxNodes);
for (n=0;n < int(maxNodes+maxNodes/100);n++) {
depthList[n]=0;
}
|
|
We'll also use a function that automatically searches for the best
available depth and returns the layer in which to place the next node.
|
|
|
Function getNextAvailableDepth
finds the best empty layer for a new Node |
|
|
function getNextAvailableDepth(z) {
// calculate best fit layer
var zd=Math.round(z * Math.floor(maxnodes/100));
while (depthList[zd]!=0) {
zd++;
if (zd > (maxnodes+maxnodes/100)) {
trace("warning: no more layers!");
return null;
}
}
// a position has been found at zd
// mark the position as taken
depthList[zd]=1;
// return the depth
return (1+maxnodes-zd);
}
|
|
Now each time we create a node, we'll know exactly where to place
it by the return value of the function above. Note that the function
may return null in the case that there are no available functions
for the depth requested.
Another convincing appearance of depth is through scale. The further
away an object is, the small it's scale. This simple relationship
works nicely with our values of z. Larger values of z (higher depths,
closer to the user) produce larger scales. A z value of zero is
the deepest a node can go.
|
|
|
A Node's size is scaled according to its
depth |
|
|
// some actionscript for scaling the node body by z
|
|
There are a few restrictions and disadvantages
to this depth ordering system. Once created, nodes can not alter their
z values. There is no vanishing point, so objects are not accurately
rendered in 3D. Concurrent depths still have a first come, first serve
layer ordering system. |
|
|
The swapDepth command is a perfectly good alternative to this approach.
However, certain situations cause objects to get stuck in swapping
loops, where no object is ever quite happy with where it is. This
CPU intensive paradox is the reason it has not been used for this
particular implementation. |
|
|
Our three dimensional node is now completely prototyped. As a final
step in bringing these objects, we will develop a method to create
them. Ideally, we would like something simple and reusable.
|
|
|
The createNode
function provides a simple and reusable method in which to instantiate
new nodes. |
|
|
function createNode(ix,iy,iz,dx,dy,dz) {
var neo="nd"+String(depth++);
var actualdepth=_root.getNextAvailableDepth(dz);
this.attachMovie("node",neo,actualdepth);
this[neo]._x=ix;
this[neo]._y=iy;
this[neo].z=iz;
this[neo].dx=dx;
this[neo].dy=dy;
this[neo].dz=dz;
return neo;
}
|
|
Note that creating a new node requires two sets of three parameters.
These parameters are all numbers.
The return value of the createNode function
is the name of the node that was created. This name is added to the
end of nodeList, an array that tracks the
names of all nodes created.
|
|
|
A node is simultaneously created and added
to the list of nodes. |
|
|
ix=random(800); // initial horizontal position
iy=random(450); // initial vertical position
iz=random(100); // initial depth
dx=_xmouse; // destination position
dy=_ymouse; // destination posittion
dz=iz; // destination depth
// create node and add to list of nodes
nodeList.push(createNode(ix,iy,iz,dx,dy,dz));
|
|
More information on the node and it's variations can be found within
the Life of the Node series of Levitated Expansions.
|
|
|
jtarbell,
July 2002 |
|