they don't seem to have any kerning pairs with space.

Because it cannot work this way. In your example “A V”, you are correct, it might benefit from adjusted spacing, but that is cause by the combination of A and V itself, not by the letters standing with a space. You cannot solve this problem by kerning SPACE and V and A and SPACE. Because those 2 single kern pairs don't know the other letter involved. Should the first letter change, than the kerning of SPACE and V would be wrong again.
You would need to take all 3 letters into account, which is hardly possible considering all the letters that can stand on each side of the space.

I've seen typefaces that kern the word space, and others with contextual kerning for select triplets (though not including the wordspace, I think).

It's pretty rare, probably because it can be dangerous, since it can result in words becoming too close (and too loose doesn't harm readability nearly as much as too tight, at least in terms of wordspace). But I wonder, what about kerning something like "f"-[space] positively? For a really... well-endowed "f" that might be good.

A more sophisticated solution though is kerning triplets (with the space in the middle) although as Ralf says it can explode.

hhp

Williams Caslon Text uses CALT to substitute different shorter-nosed f's before spaces, depending on what the letters are following the space. I also negatively kerned some of the space-letter combinations, like space V, space T. I think the triples involving the substituted f's work pretty nicely. This idea came from Charles Ellertson, if I remember rightly.

Certain digital fonts used to have letter-space kerning originally, but have been revised to remove it, e.g. Gill Sans, IIRC.

Myriad has some space-letter kerning, but not the corresponding letter-space kerning.

Hrant: But I wonder, what about kerning something like "f"-[space] positively?

Yes. This is more important than tightening word spacing for letters like A and V. If those are well-spaced to begin with, then the visual impact on word spacing isn't very great. But an f that extends a long way into the word space will come very close indeed to a following cap or ascender, especially a T.

This (above) is the f-to-space kerning in Brill Roman, with the same sequence without kerning below. The value here is +45/1000 units.

Oh, I remember this came up last year, and I did an example with "of Two", which is here:

The top is as the font works now, with a shorter nosed f substituted and the space-T kerned. The middle is with the long nosed f with the space-T kerned. The bottom is the long-nosed f and no kerning on space-T.

I think this works better than just kerning out the f-space combination. But I don't think the whole issue is that consequential.

I wasn't going to comment, but...

You need to be very careful kerning anything with a space. It is a slippery slope, and easy to get in trouble.

I've used this example before on Typophile. Take Adobe Jensen. Set f_wordspace_quotedbleleft_a

The wordspace just vanishes -- esp if you fill out the words, like

...variance of "allowed" tolerances...

Hugh Ambrose's book, The Pacific, is full of these. The cause of this problem with Jensen isn't so much the "f" but the kerning of the quotes back into the wordspace. The two kerns in concert kill the wordspace.

If you want to kern an f with a wordspace, be sure to test it when followed by a few capitals -- capital J is usually the worst, as in "of Jewish origin." Some lower-case letters can give problems, too (b,h,i,j,k,l), as well as the left quotes.

I prefer to create a special f, where, when followed by a wordspace, the flag (nose) of the f is shorter, and the sidebearing a bit greater -- though I suppose increasing the sidebearing is the same as using a blanket kern. I believe this is what Bill showed with his Caslon. With OpenType, a calt substitution can make its use automatic.

As others have said, the real answer, in so far as there is one, comes from the design and spacing of all the letters, and the compromises they occasion. Pairs kerning offers only limited help. You have to set it up for the worst case. Contextual kerning can improve greatly on that.

Here is the mental model I use for contextual kerning. Back in the days of hand-set type and the composing stick, you set the line, then added more space to the word spaces to drive the line out to full measure. If a particular space was optically larger, you just didn't add as much space there.

The problem with machine set type, esp. computers, is that adding equal space to drive out the line is automatically applied everywhere. The "programmatic handwork" then, comes in taking out space -- kerning. Essentially, that requires triplets. For example, c_space_w usually requires less space than h_space_n. for optically even spacing. Etc.

If you can put this in the font and your layout program supports it, fine. Otherwise, easy enough to write a script to put the kerns in with a layout program -- InDesign, TeX, whatever. If you use regular expressions, you can get quite close to the thinking/approach of class-based kerning. We've done this in our shop since the late 1980s, when we drove a photocomp machine (Linotron-202) with TeX.

But for kerning between just two letters, don't go beyond what the worst situation allows when you add that third letter -- e.g., f_space_J, f_space_quoteleft, etc.

Bill, Charles, I've opined this before, and I still think making the "f" more modest (compromising the bouma) before a space is backwards: the beak of the "f" (the only worthwhile ascender in the Latin alphabet) is kept at bay mostly to avoid a collision, and the blank space totally eliminates that problem. So -assuming you're not getting into kerning triplets- I think kerning the "f" positively with the space is generally the best compromise; I might even contextually make the beak bigger. Remember, too much space isn't nearly as bad as too little.

hhp

Well, effectively substituting different f's depending on the triplet amounts to triple kerning. I used three different terminal f's. The regular terminal was this with the shortened nose (red-terminal, black-regular) :

and one alternative was this, same side bearings with a longer bar to the right:

And then there was another one like the above, but with wider side bearings on the right. I substituted one of the three using CALT depending what seemed to work best with the triplet.

> You would need to take all 3 letters into account, which is hardly possible considering all the letters that can stand on each side of the space.

“All the letters that can stand on each side of the space” come in exactly the same count as “all the letters that can stand on each side of the gap between two letter”. So, in the worst-case scenario, you need to double the size of the kerning table. (This ignores the fact that some pairs may be not kerned correct, as in f+space+opendblquote mentioned before — usually, there is little need to kern f+opendblquote. So it has a chance to become more than 2× larger…)

usually, there is little need to kern f+opendblquote.

Assuming that by “opendblquote” you mean the quotedblleft (U+201C), then this statement is true only if one does not intend the font to be used for German.

Right, I completely missed this. And shame on me, since I suggest to my users to remember this :-[ — and recommend seeing http://en.wikipedia.org/wiki/Non-English_usage_of_quotation_marks.

— So it may happen that the limit is indeed about 2× number of non-spaced kerning pairs. BTW, can somebody see why just using the “usual” kerning even when separated by space won’t solve the original problem?