Marco, I agree. I did some basic design work on an Ethiopian system and it was decided to follow the same implementation system as Thai. We don't encode every possible Thai glyph.
We felt that if it were ever Unicode encoded we needed to use the decomposed characters rather than decomposing them at translation time. Decomposed characters also eliminated a custom IME. It would be implemented in the font engine. Unicode at that time had not allocated more than a single 256 byte chunk for any single script other than Unihan. Besides it could be implemented as a single byte code page. There are only a few invalid combinations and we could not get any consensus on which ones they were because of regional differences and some had become obsolete. Carl
