Peter Schmitt writes:

 > Yes, colour is an attribute of the font
 >      and therefore it should be handled as such.

i think this is an oversimplification. color is an attribute of a region; if
you limit your vision then perhaps an attribute of rendered shapes, but fonts
are only one type of objects that belong to this category. what about graphical
objects like lines etc, what about background, ...

i like to think of color being an attribute of the region you render objects
in (and the background is then a third dimension extension to that region
concept) -- none of that is supported in TeX (or eTeX or Omega) as of now but
that is what i think one has to look at for models

 > Switching the colour of a font is logically
 >      the same as switching to another font.
 > A font change is efficiently handled by TeX and needs no hack at all,
 >   and for a coloured version of a font a virtual font could be used.

well, technically you can do that for coloring glyphs and it is certainly
better than using minimal size changes to signal attributes. however both
methods are more or less impractical in large scale usage (beside being a
hack)  and i don't think they implement the right model

i'm just about to finish TLC2 which is printed in two color, in total it uses
about 200 fonts (tfms) given different sizes special show cases math etc etc,
a lot of them are used in black *and* in the second color. so just maintaining
a full set of additional tfms to signal color would be a nightmare, despite
that we use the color in examples for rules, picture graphics, and background