Remove unneeded pairs.

[schriftgestalt]

First, thanks for sharing this.

I think the list can be greatly improved by removing pairs that don’t need kerning or can be covered by class kerning:

1. Pairs with space. The first 30 pairs are mostly with space. Those are usually not kerned.
2. Most pairs with accented letters. Most Uppercase with accents need the same kerning as the base glyph. This is true for A lot of the lowercase, too. Exceptions are T|V|W|f|+letter with top accents and pairs with quotes.

[andre-fuchs]

Thanks Georg for your suggestions. Your second point makes total sense. I should implement groups directly for the uppercase letters. Suggested character groups in addition to a filtered list would be the way to go, I feel. Regarding lowercase letters, this might need more revision of the whole list, I expect. Do you want to help me with these lowercase groups?

I disagree with you on the first point, though. The significance to kern letter pairs containing space is confirmed by this study, I interpret. The values required depend on the design of the characters, of course. But these pairs occur exponentially more often. The font Univers ("Pro" version by Linotype) is a famous example of overdoing it partly (I remember the letter combination of space + T, for example). But I think, generally speaking, the rhythm can be improved by taking these pairs into consideration as well.

[schriftgestalt]

For the lowercase and lowercase+accent pairs, you can use the side-bearing structure info. You don’t need pairs with top accents, and the other letter has nothing above the x-height.

[andre-fuchs]

Thanks Georg. What do you think how important traditional kern tables are today? The cat would bite its tail here. Group-based kerning generates a lot of redundant kerning pairs, as the evaluation of kerning tables has shown (see my first point Popular Kerning Pairs).

I think it would be helpful to make recommendations about these groups and the letter pairs filtered with them. In the end, however, one should be aware of the fact that this is a very long tail of redundant combinations. Don’t you think?

[schriftgestalt]

I don’t think you need to consider kern tables any more. I only ever do groups based kerning. If one needs a kern table for whatever reason, he needs to find a way to filter it down to fit.