The reason for this is that there are two kerning tables, and the older type ('kern') has a limited fixed size (~50,000 pairs). There is not enough room to have every character vs. every character in the first table, so just the basic Latin-based kerns are entered into the first table.
The second table uses a class-based system and can contain the whole-set.
On einteresting research project might be trying to cram the largest possible number of kerns into the first table (depending on available space), based on some statistical analysis of kern-use frequency/visual impact of the full kerning set.
Hello Petr,
The reason for this is that there are two kerning tables, and the older type ('kern') has a limited fixed size (~50,000 pairs). There is not enough room to have every character vs. every character in the first table, so just the basic Latin-based kerns are entered into the first table.
The second table uses a class-based system and can contain the whole-set.
On einteresting research project might be trying to cram the largest possible number of kerns into the first table (depending on available space), based on some statistical analysis of kern-use frequency/visual impact of the full kerning set.