Syntax highlighting doesn't necessarily have to work at the lowest granularity, though; for some uses, merely distinguishing between comments and non-comments is acceptable, and that's still 'syntax highlighting'. Of course, as you point out, true 100% syntax highlighting needs to fully parse the entire program; why not do that, though? I guess it would be too computationally expensive for certain sizes of program, but it could still update in near-realtime, no?