Actually:

• He had already published the first editions of Volume 1, 2, 3, and the second edition of Volume 1, by 1973. It was in 1977 when the publishers sent him galley proofs for the second edition of Volume 2, having switched to phototypesetting (away from hot-metal typesetting a la Linotype, though IIRC it was actually Monotype) that he was disappointed with the results. And he had some back-and-forth with them and they did improve their fonts (https://tex.stackexchange.com/a/367133/48), but he was still dissatisfied.

> I didn't know what to do. I had spent 15 years writing those books, but if they were going to look awful I didn't want to write any more.

• At this time he came to know of the existence of digital typesetters. Typesetting with computers had existed before, but it had always seemed a crude toy, rather than something suitable for “real books”. But he saw Patrick Winston's Artificial Intelligence that had been just published (I think he got an early proof copy to review or something), and he realized for the first time that digital typesetting was an option (apparently Winston's book was printed at >1000dpi, and Knuth later got his hands on a machine that claimed a resolution of 5333 dpi: see this wonderful comment from Knuth's student and “right-hand man”, David Fuchs: https://news.ycombinator.com/item?id=20009875)

• In fact it was the fonts that he was dissatisfied with rather than the typesetting, so METAFONT was in some sense the primary/motivating project and TeX was only written in order to be able to use METAFONT.

• Actually his first idea was to simply take the old fonts, get high-resolution scans of them (not easy to obtain at that time) and use them directly. He approached Xerox Research Center but:

> I asked if I could use Xerox's lab facilities to create my fonts. The answer was yes, but there was a catch: Xerox insisted on all rights to the use of any fonts that I developed with their equipment. Of course that was their privilege, but such a deal was unacceptable to me: A mathematical formula should never be "owned" by anybody! Mathematics belongs to God.

• So he went home and (after trying a bit with TV cameras) tried projecting photographs of the pages onto the wall and tracing the outlines, and it was while staring at these images that he realized that the shapes of letters were not arbitrary but there was some logic to them (e.g. in the font he was using, the spacing between the vertical strokes in 'm' was equal, and equal to that in 'n'), and he decided (as a computer programmer) to capture this design in code — something that had never before been done. The hardest letter to capture this way is S, hence the paper in the OP.

> Finally, a simple thought struck me. Those letters were designed by people. If I could understand what those people had in their minds when they were drawing the letters, then I could program a computer to carry out the same ideas. Instead of merely copying the form of the letters my new goal was therefore to copy the intelligence underlying that form. I decided to learn what type designers knew, and to teach that knowledge to a computer.

• This is also why METAFONT never really caught on among typographers: as Charles Bigelow (quoted by Richard Southall, https://luc.devroye.org/Southall-METAFONT1986.pdf) observed, “the designer thinks with images, not about images”. Knuth did not want crude “geometric” constructions of letters (as some prior 16th century typographers had attempted: https://www.ams.org/journals/bull/1979-01-02/S0273-0979-1979... and as some typographers only passingly familiar with METAFONT think!). He wanted actual real typographically beautiful shapes, but to be able to generate those shapes with code. This is obviously much harder than simply drawing the shapes using visual intuition, even if it enables variation. (See “The Concept of a Meta-Font”: https://gwern.net/doc/design/typography/1982-knuth.pdf — again, many people in the typography world confuse the abstract concept of a meta-font introduced in this paper with (their incorrect impressions of) the METAFONT program, and omit crediting Knuth for variable fonts).

• The second edition of Volume 2 was not printed with Linotype. Yes the machines still existed in Europe and he talked to typesetters (he mentions in particular a person from Belfast), but it was in fact published using TeX (the first version, TeX77 and MF78). He was still unhappy with the results, though, and spent a few more years learning more about typography and working with people like Bigelow and Hermann Zapf, before the rewrite into the current TeX82 and MF84 (and current version of Computer Modern). I think it's only with the third edition (1997) that he's finally satisfied.

Not so about the Linotype. Back in 1980, I personally ran the Alphatype CRS phototypesetter (bought by DEK for the purpose), in the basement of Margaret Jacks Hall, that produced the entire camera-ready copy of The Art Of Computer Programming, Volume II, Second Edition. The DVI files and Computer Modern fonts were created by the early, Sail-language, 36-bit versions of TeX and Metafont that were later redesigned and implemented to be more cross-platform. Knuth rewrote the firmware that resided on the Alphatype (in 8080 assembly language), and I wrote the code that translated from DVI and drove it from the DEC20 mainframe over a serial line (trickier than it sounds; see our joint paper "Optimal prepaging and font caching" ACM TOPLAS Vol 7 Issue 1).

I just want to add that they are a gorgeous set of books and I am so happy he did this. While a good chunk of the content is above my pay grade it is still enjoyable to flip through them and read about things like MIX. Gorgeous typesetting. And his writing is so very engaging for such a dense topic.

Note that they weren't directly printed with a hot metal casting machine (I believe Monotype was used, since it output discrete letters which did not need to be sawn apart as the Line of Type the competitor used), but rather what could be composed on a hot metal casting machine was, then additional spacing material and special characters and extensions which weren't available from that keyboard were sourced and the whole put together as a composed galley, then a proof was pulled and once approved, printed, then photographed to make a negative which was then used to make an offset plate for actually printing the book.

Suddenly, I don't feel so bad about my own procrastinitis.

For the work on Euler, this article¹ (https://www.sciencedirect.com/science/article/pii/S240587262...) goes into the whole Digital Typography program at Stanford for whom, one of the projects was creating those outlines for the Euler math fonts. It’s worth remembering that at the time, not only was Metafont the only outline-based font technology,² but things like scanners were rare to nonexistent and the bitmaps that were used to determine coordinates for the curves of the fonts were hand-drawn on fine-lined graph paper (and sent to Hermann Zapf for approval).

⸻

1. Funnily enough this is the second time in two days that I’ve shared this article, albeit in different contexts.

2. As far as I know, although I could be wrong.

Maybe another old example of this scanner bug?

https://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres...

To spend nine pages full of mathematical formulae just to write a single letter (more) nicely shows the rigor/perfectionism/OCD that is the hallmark of Donald Ervin Knuth.

Thanks for giving us beautiful layout and better-looking fonts. Every time I write a new paper when I press "compile" in Overleaf I'm greatful that he made our work more beautiful, and it motivates me to make the content matter, too.

How would Knuth opine on consummate Vs, I wonder...

Trogdor!

Among the uppercase letters, "S" is the only one that's not either based on a single primary ellipse, or combined with straight lines that provide structural constraints. The entire glyph is governed by continuously changing curvature with no stabilizing axis or primitive shape to enforce proportions.

This results in a more complex and less obvious mathematical definition.

Also, a naively symmetrical "S" tends not to look good, probably because of these same issues, so needs further adjustment to match our visual expectations. This complicates the definition further.

Added. Thanks!