Peter T. Daniels wrote:

> [..] encoding Ethiopic and Cherokee as if they were katakana [...]
>
> Cherokee _is_ -- you must've meant Cree.

Yes, sorry.

> But do you want your typography subsystem to have to rotate
> characters?

That would rather be implemented as a *ligature* between a "consonant" code
and "vowel" code. The facility to map a sequence of codes to a single
"ligature" glyph is already in place in modern digital fonts (e.g.
OpenType).

E.g., the Arabic laam+alif ligature or the Latin f+f+i ligature are encoded
with a code for each letter: the ligature occurs automatically during
display.

> Rotating seems to be a very memory-intensive procedure.

Not so much: if I recall correctly, it only involves a few matrix operations
on the coordinates of the control point.
However, algorithmically rotating consonants wouldn't probably obtain
typographically correct glyphs.

_ Marco