1. Single line of text of a given size/position/orientation in the 3D world,
in a default font. Useful for labelling objects (anything else?).
2. A VRML equivalent of Inventor's Text3 object-- multiple lines of text,
unformatted, left/right/center justification (without the arbitrary extrusion
feature).
3. Multi-line, formatted text, specified as font size, width of rectangle to
format into, and position/orientation of the rectangle in the 3D world.
4. Multiple fonts/styles for 1 and/or 2.
5. Internationalized (non-ASCII) text-- Unicode?? (does HTML do something
here?)
6. Arbitrary HTML (with inline images, hyperlinks, the whole 11 yards)
formatted into a rectangle in the 3D world (actually, we'd probably want to
limit it to the HTML stuff that can appear between <BODY> and </BODY>...).
All of these can be faked using the current VRML spec, either by generating a
ton of polygons or generating texture-mapped rectangles. I agree, a better
way of specifying text is needed to save network bandwidth when transferring
VRML scenes, save disk space on servers, and to make authoring easier.
Screen-space objects cause lots of implementation headaches, I'd strongly
suggest leaving out 2D (screen-aligned or always-on-top) text.
Inventor's Text3 can extrude itself along arbitrary (linear or NURBS)
profiles, which is a cool feature but which is certainly not a critical
feature and which relies on having outlines available. I suggest keeping the
specification simple enough that low-end machines can just use simple stroke
fonts instead of requiring text to be rendered as lots of polygons.