Kerning pair bloat?

Cory Maylett

I've been getting feedback from the review team at a leading type distributor. Although they've accepted the font family, they've also recommended additional kerning pairs to address increasingly minor spacing issues, then provided examples of the kinds of minor kerning issues they're seeing.

There are around 700 glyphs in each font, which contains Latin, Greek and Cyrillic characters. The level of kerning they've suggested, when carried across an entire font in the family, amounts to around 6,000 kerning pairs each, and this is after fine-tuning the sidebearing.

Six-thousand kerning pairs per font seems a bit excessive, but I've gone ahead and spent the past couple of months creating them over weekends. I've also noticed a growing trend in recent fonts toward more and more kerning pairs that extend way beyond the usual awkward-fitting combinations to include nearly every possible pairing.

I suppose computer processing power has increased to the point of being able to handle the extra kerning, but I'm still concerned about bandwidth load for web fonts. As a website designer myself, I'm a little hesitant to embed 150 KB WOFF files.

I've considered breaking the web fonts into Latin, Cyrillic and Greek subsets, but after pouring through online Greek and Cyrillic publications, it seems there's a whole lot of Latin characters used in them, which probably rules out Greek or Cyrillic without including at least the Latin Extended-A block in each.

Anyway, I'm wondering what others might think of these large numbers of kerning pairs. Every font's kerning needs is different, of course, but at what fuzzy, hard-to-determine point do the drawbacks of increased bloat begin to exceed diminishing benefits. Also, if anyone has any insight or opinions on the value of subsetting web fonts, I'd appreciate reading it.

James Puckett

.

Nina Stössinger

Cory Maylett said:

The level of kerning they've suggested, when carried across an entire font in the family, amounts to around 6,000 kerning pairs each

I may be pointing out the obvious here but are you sure your kerning groups are optimally set up? I mean, outside of the fundamental point you raise, you may be able to save rather significantly on pairs that way.

Cory Maylett

Thanks Nina! The fonts started out with well-organized kerning classes, but after all the fiddling with exceptions over the past few weeks, I'm sure there's room for some optimization.

FontLab's automatic font class generation feature consolidates it all into about 400 fewer kerning pers per font than the roughly 6,000 I have now, but it introduces other kerning issues that I'd need to identify and deal with one at a time. Multiply this by the ten fonts in the family, and my brain begins to go numb.

Despite the 900-page FontLab manual, I'm still finding it short on detail describing exactly what some of its functions actually do.

Chris Lozos

I find that any "automatic" font pair generation command tends to error on the excessive [and often strange] side.  Carefully look at those as well as pair down exceptions to what is really needed.