John, you seem to say "normalization" but mean "decomposition". Please note that there are several normalization forms, and the most popular one is NFC, typically using code points for precomposed characters. Your email suggests that MacOS is using NFD, which I find surprising.
On the issue of string comparison, I implemented a function last week for ICU 2.2 (this summer) that compares strings efficiently, decomposing only where necessary (i.e., incrementally and only where there is a difference between the strings). The same function also has options for case-insensitive comparison and code point order. On the general issue of fonts vs. various canonically equivalent forms of "the same text", it seems like a layout engine could automatically select one form if it does not find glyphs for the other. I am not an expert on fonts and layout engines though. markus
