It depends how good you are at setting up classes.

edit... and how well spaced the font is before starting kerning.

How much does an automobile cost?

I've never counted them.
You could, either by opening up the fonts in a font editor, or in Quark XPress' kern table editor.

4872.

…so far.

Neue Helvetica (the Mac system version) has around 100 pairs per fonts. Surprisingly low. What exactly do you mean by »Roman alphabet«, just A-Z, a-z, no accented letters, no punctuation?

Williams Caslon Text over the four styles has a total of over 16,000 kerning pairs and over 600 kerning groups. And it is currently in use without any kerning at all!

Thank you Russell, Jens and William (who responded while I was writing this!) for your feedback.

For this discussion, Roman = not Cyrillic, Hebrew, Chinese &c. We define below as a complete alphabet: a simple regular, shift, option and shift+option for an American English Macintosh keyboard:

`1234567890-=
qwertyuiop[]\
asdfghjkl;'
zxcvbnm,./

~!@#$%^&*()_+
QWERTYUIOP{}|
ASDFGHJKL:"
ZXCVBNM<>?

`¡™£¢∞§¶•ªº–≠
œ∑´®†¥¨ˆøπ“‘«
åß∂ƒ©˙∆˚¬…æ
Ω≈ç√∫˜µ≤≥÷

`⁄€‹›fifl‡°·‚—±
Œ„´‰ˇÁ¨ˆØ∏”’»
ÅÍÎÏ˝ÓÔÒÚÆ
¸˛Ç◊ı˜Â¯˘¿

To rephrase the question: "For the glyphs shown above, on average, how many individual kerning pairs (not kerning classes) would a type designer working for a company such as FontShop producing a typeface of similar quality to that of FF Meta OT Book (just for the glyphs shown above, not all 524) how many individual kerning pairs would their be?

My question about how much an automobile costs was not frivolous; the answer is "it all depends." The design of the individual characters will dictate how much kerning will be needed. Oswald Cooper designed his eponymous black face to be cast in metal, without the ability to kern as we know it. Consequently, he designed his letters in such a way that kerning--even if it could be accomplished--was unnecessary.

@ Dean: ...on average...

I concur with Nick, it depends.
I have made some fonts with a lot of kern pairs, others with not many.
And my "best practice" has changed over the years.

So if you want to know what the figure is "on average", you will have to do the calculations yourself, because unless someone actually sits down and does that donkey work, all you have is guesswork, which even if educated, is a shot in the dark.

Why does this matter?

For a sans with upper & lowercase + kerning classes, I tend to hit in the 320-380 range. I hadn't compared or even noticed the number of pairs before. I was surprised to see how the numbers end up pretty close. A square sans is about 220. A superelliptical, like DDT is about 280. My earlier fonts had more pairs, probably because my spacing skills weren't as good. If you eliminate kerning classes, the answer is easy: exactly 2048 pairs. Some all caps fonts come in at less but everything else is 2048.

If you’re talking about the Mac Roman char set and no kerning classes, then you’re essentially asking what is the average number of kerning pairs in a legacy Postscript Type 1 font.

Still, it depends. It depends upon the style of the design, the effort put into shaping and spacing to obviate kerns (not necessarily a good or bad thing, just a design intent), whether the design is intended for display or text, and the extent to which a designer chose to go (kerning accented chars or not, which punctuation combinations to give attention to, etc.)

In the original Whitman text fonts, there were around 1800 individual kern pairs. In the Display, more like 5000.

I once did an analysis of some rough breakdowns. I can’t find the Whitman analysis; but there is this old thread, with data from a few other fonts:

http://www.typophile.com/node/459#comment-2948

Unfortunately, many posts from that era end up getting borked in the archive (whenever an accented character occurs) and there is some long post that I made the day before this one that seems to have had some really good information (according to the succeeding comments) but which got truncated. I can’t remember what I must have said.

By the way, the reason there are so many kerns in Williams Caslon is because its glyph set is very large, including CE characters, Bold Italic small caps, swash characters, and 6 styles of figures. You have to do "exceptions" for the accents, and the swash are kerned to both lower case and small caps. So everything explodes exponentially with more characters that need to be kerned with one another.

Also the italic follows the varied slants of the original (though all is made 3% more upright). I think because of the varied slants, it took a lot more kerning to get the evenness of a modern italic. There are about double the kerns in the italic as the roman.

Also this follows the FB method of not putting negative side bearings in vwxy, so that they won't touch even if kerning isn't operating. This creates a need for more kerning to get the diagonal characters even in the final.

Bill,

Who cares?

>Who cares?

Evidently, Christopher, since he asked.

Christopher, I don't know how many of the kerns were for just the Mac roman set, as my software just tells the total for each style. For just the Mac set, no open type options, it would be way less.

Oh Bill, you are so predictable!

I am trying to develop a sense of the amount of manual labour that goes into establishing good letter-spacing for a typeface.

I am trying to develop a sense of the amount of manual labour that goes into establishing good letter-spacing for a typeface.

One could argue that putting more effort into spacing reduces kerning time. Especially in fonts that do not have kerns. And the number of kerning pairs in a font dot not necessarily correlate with how much time was spent kerning it.

One could also argue that most folks doing this work today don't use pairs, they use classes.

Cheers,

T

...the amount of manual labour...

I have no idea how much time I spend on this or any other aspect of font making.
I didn't get into this line of work to become a farking assembly-line robot.

^What they said… lol

While we're on the topic: what's the technical limit for the number of kerning pairs in an OTF? When I first started making OTFs about 7 years ago, I didn't set a limit when expanding class based kerning. Some fonts would expand to about 10 kilopairs. When testing in Adobe CS1 apps, the kerning didn't work at all. So I capped the limit to a couple of kilopairs and all was well. What is to actual breaking point for kerning pairs in an OTF?

Why would you expand the kerning for OTF when you can use class-based kerning?

Some apps don't support class based kerning. So even basic stuff like commas, emdashes, ellipsis, base quotes and smart quotes would end up unkerned in those apps if the class kerning wasn't at least somewhat expanded. For all caps fonts, unexpanded kerning meant the "lowercase" set ended up with no kerning at all. For the first few years of OTF, MyFonts samples didn't display class based kerning. If you didn't expand the kerning, I'd only see the class parents kerned.

Now, with web embedding, class kerning isn't fully supported so, if I don't expand kerning, it won't kern.

In earlier OTF times, there was an OS9 app that supported OTF but not kerning classes. I can't remember which one but a few customers mentioned it so I made sure I expanded everything.

In Windows there was something . . . Corel Draw? I think there was a version that supported OTF but not classes.

what's the technical limit for the number of kerning pairs in an OTF?

There's no simple number. The format is limited by the size of a subtable (64kb), though we now have the "extension" mechanism to work around that, in newer environments. Some individual applications will start to misbehave if some limit is exceeded, but the nature of that misbehavior and the size of the limit can vary by app and even by app version. Some applications have no limit whatsoever (for example, InDesign).

Examples of different kinds of misbehavior:

- only recognizing the first n pairs
- only recognizing the last n pairs
- the app crashes
- the app fails to recognize any kerning in the font

Generally, Adobe notified all the application developers whose apps they saw misbehaving, about 8-10 years ago. I know that Framemaker and QuarkXPress improved.

Regards,

T

I do as many as need to be done. I don't count--that would just take more time than it is worth. The pairs involving diacritics take the most time because there are more variables that simpler classes can't handle.

Basically, rather than count how many stones you roll up the hill, you just do it and move on.