Metafont – ?

Andreas Stötzner · July 2022

I had never anything to do with Metafont and I wonder if anyone of you could enlighten me about its actual status, importance, merits, prospects etcª in present day typographic practice.

On Wikipedia I read: “…the system has not been widely adopted by professional type designers…”;

however, it may be that someone who is dealing with special subjects or scientific editing may have some experience to share.

John Savard · July 2022

Donald Knuth is very well respected for his achievements in computers and mathematics. This does not make him a typographer.

The original Computer Modern typeface was created by means of Metafont.

Of course, it could be argued that, since Donald Knuth was not a professional type designer, the faults of Computer Modern are due to that, and not to anything wrong with Metafont.

While Donald Knuth's TeX also differs from ordinary typesetting markup languages, that didn't prevent it from being very well recieved by mathematicians, as it covers the typesetting of mathematics very well, in a way that nothing else does.

But the basic principles of Metafont are built around imitating the motion and behavior of the pen. And it doesn't seem to do that quite well enough for one thing, and for another that's not the kind of tool you want for every typeface. So unlike TeX, Metafont just didn't appeal to type designers - even back in the days when it didn't have to compete with the entrenched True Type/Open Type standard.

So it's basically ignored and irrelevant in present-day typographic practice. Apparently for good reason, but of course that does not mean that it couldn't have more merit than it appears to on the surface - and that some good ideas might be found by going back and studying it.

I've found the following page

https://www.yld.io/blog/lessons-from-metafont/

which discusses the matter in more detail.

And I learned some important things from it that I did not know. I thought the new, improved version of Computer Modern in current use was designed by someone else using conventional tools like TrueType. But that is not the case. Instead, Hermann Zapf himself helped Donald Knuth improve Computer Modern, and improve Metafont.

Given that, I now doubt that Metafont, in its improved version, is as bad as I thought.

But while it is now possible to design attractive typefaces with it, that's still only attractive new typefaces. Because it is pen-based and not outline-based, it is not the tool to use if you want to make a digital version of a typeface that previously existed in metal.

To me, that seems like a fatal weakness. (And in its new version, it may even be possible that someone could, as a tour de force, make an imitation of Times Roman in Metafont; it may no longer be impossible, it may just be doing it the hard way.)

Andreas Stötzner · July 2022

thank you John, this is quite what I suspected.

thanks also for the worthwhile further reading link.

Michel Boyer · July 2022

It might also be interesting to have a look at previous posts on the subject on typedrawers https://typedrawers.com/search?Search=metafont

John Savard · July 2022

Although I didn't find too much in the other threads that search turned up, I did find this one quote:

Dave Crossland said:

James, Simon and Dan are partially right; the AMS Euler fonts were designed by Zapf on paper and digitised by students at Stanford under Knuth's supervision, but they quickly gave up on using METAFONT as Knuth intended, and used it in an Ikarus-like way to draw the outlines Zapf had sketched.

which doesn't look good for Metafont. Although it's not clear to me how Metafont can be used to draw outlines. Outlines that will be filled when printed, that is.

Mark Simonson · July 2022

I think the failure of Metafont is that the model it's based on is too simple and not flexible enough for creating (or re-creating) typefaces. It always struck me that Metafont fonts look as if they were drawn with a Speedball pen. That's kind of what the model is.

Type, from the very start, was created by cutting the shape from a piece of metal, where any arbitrary shape is possible, not just shapes you can make with a pen. The reason the outline-based font has prevailed with digital fonts is because it has the same kind of flexibility with regard to making shapes.

Simon Cozens · July 2022

I think there's more to say on this: METAFONT is both too simple and way too complicated.

The first thing to note is that Donald Knuth is a genius, and like many geniuses, came up with brilliant ways to solve problems in different disciplines that were way ahead of their time. The problem with solving things way ahead of your time is that there's little else out there to compare with. So you make a lot of progress and you blaze a lot of trails, but you also make what look like mistakes with the benefit of hindsight. (And the problem with Donald Knuth is that he has unwittingly gathered a fanbase who cannot conceive that any of his ideas actually turn out to be mistakes.)

So DVI, TeX's page description language, was a great idea at the time but nobody uses DVI today. Instead, they used PostScript and now PDF. One of the first ways that people extended TeX was to make it use PS, and then PDF, instead of DVI. And similarly for the text encoding system, to allow the use of Unicode, and similarly for the font system, to allow the use of Type1 and eventually TTF fonts instead of METAFONTs. Knuth's fans don't like me saying this but the entire history of TeX for the past forty years has been one of replacing Knuth's trailblazing technologies with standard ones that people actually use instead.

It's not true that the METAFONT system is limited to defining glyphs based on pen shapes. You can position "left" and "right" points of an outline independently. But I think it is fair to say that most of the examples are based on tracing the centre points of a skeleton and having METAFONT fill in the left and right points, and the system does reward working in this way.

The bigger problem is that, as Knuth says himself in Digital Typography somewhere, METAFONT is essentially a constraint solving system which draws glyphs by solving simultaneous equations, and that isn't really the way that designers think about glyph drawing. Again, it's possible (as the "Nara" font example shows) to use METAFONT without the equation solving style, but again the system rewards it.

Fundamentally as a mathematician, Knuth saw type design as a problem to be solved using mathematics. I think he imagined that fonts would be made by programmers sitting down alongside designers, translating their design ideas into equations to be solved, but I think he was also realistic about how unlikely that was to happen. (And of course this was all before the widespread use of graphical user interfaces and interactive computing, so it wasn't a time when you could easily conceive of designers being the primary operators for computers.) The irony of course is that today we have font engineers who are programmers sitting down alongside designers making fonts together...

John Savard · July 2022

Simon Cozens said:

(And the problem with Donald Knuth is that he has unwittingly gathered a fanbase who cannot conceive that any of his ideas actually turn out to be mistakes.)

But how much of that fanbase is actually still around these days? I had been looking for it, and an archive of TUGboat turned up in my Googling, so that I could have given links to a less jaundiced view of Metafont than mine.

Simon Cozens said:

It's not true that the METAFONT system is limited to defining glyphs based on pen shapes. You can position "left" and "right" points of an outline independently. But I think it is fair to say that most of the examples are based on tracing the centre points of a skeleton and having METAFONT fill in the left and right points, and the system does reward working in this way.

It is good to hear that Metafont does include a way to define fonts as outlines. My concern would be that one of the "rewards" for doing it the preferred way... is gaining access to the parametric font goodness of Metafont. If you lose that when you step out of doing things according to the standard Metafont model, then one of the main things Metafont has to offer is lost.

Simon Cozens said:

Fundamentally as a mathematician, Knuth saw type design as a problem to be solved using mathematics.

I'm afraid that I can't give you any points for this as a brilliant insight. Knuth said as much explicitly in his original paper on TeX and Metafont.

And now to go back, and take a quote from an earlier part of your post out of order:

Simon Cozens said:

So you make a lot of progress and you blaze a lot of trails, but you also make what look like mistakes with the benefit of hindsight.

One could, of course, claim that seeing "type design as a problem to be solved using mathematics" is a mistake that didn't need hindsight to see. But the context in which Knuth made the remark is important too. Typography, reduced to its ultimate goal, is the art of specifying that certain areas on a page will be blackened by ink. Such a specification can be made mathematically, and so in that sense, it reduces to a mathematical problem.

Knuth certainly was aware of the fallacy of too much reductionism. This is why Metafont had a pen-nib model in it; he was attempting to choose a particular mathematical model that would meet traditional artistic typeface designers halfway.

And, of course, that's where a mistake was made that only became glaring once hindsight was available.

It certainly makes sense to add compatibility with existing standards like Unicode, PostScript, and TrueType to TeX. But it would also make sense to take Knuth's trailblazing work, and modify it to correct some of what are now seen, in hindsight, as mistakes - to add more flexibility. Apparently some of that has been done, but not enough to make Knuth's work a viable competitor to the current standard technologies. Nor have the current standard technologies, for their part, stolen all of Knuth's good ideas. That is what I see as a sad missed opportunity.

Hin-Tak Leung · July 2022

While I agree with most of what @Simon Cozens said about Metafont, it is perhaps written with a somewhat unkind point of view.

Yes, I am a Knuth fan boy

. Knuth's work on Metafont/TeX was created in a era before WYSIWYG (what you see is what you get...) and GUI. It is also based around a clear separation of content vs presentation. People might be happier with fonts being outlines these days, but what's wrong with fonts being strokes, similar to hand writings? If history had gone the other way, with fonts being strokes, variable fonts (Metafont was perhaps a pioneer in that direction), text shaping - strokes connecting across glyphs - might have been implemented easier technologically, instead of the lookup-upon-lookup in opentype with outlines.

I agree some of the aproaches are evolutionarily dead ends, and outdated, from the type designer's point of view. That said, I think one should understand his work in the context that the role of "type designers" didn't/doesn't exist: making a documents/texts visually pleasing is not a very interesting work, and rather tedious, and therefore should be parameterized and automated by mathematics.

Andreas Stötzner · July 2022

Hin-Tak Leung :

…
that the role of "type designers" didn't/doesn't exist: making a documents/texts visually pleasing is not a very interesting work, and rather tedious, and therefore should be parameterized and automated by mathematics.

sorry, but this is nonsense. It seems you mix up at least two different things.

1st, the role of the type designer is different from that of the editorial designer (or who ever creates a page of text).

2nd, a type designer creates a typeface which later becomes a material / tool for setting text.

3rd, making a document/text visually pleasing is in fact a very interesting work. For those who have a sense about the power and possibilities of decent typographic text design.

4th, even in Knuth’s days the role of a type designer existed. And he knew about that.

The idea of replacing a font makers work (or an editorial designers choices) by mathematics and automation, is as clever as the idea to let mathematics rule the preparation of tasty meals or Mozart compositions. I guess Knuth himself was aware of this, more or less; but can the same be said about his followers?

Dave Crossland · July 2022

To be fair to hintak, I think he's not necessarily did that is what he believes, but that this belief is embedded in the ideology of Knuth' software. Which I think is correct; many college/grad-school students have told me "latex makes my documents beautiful, so I don't need to learn about typographic design"

Dave Crossland · July 2022

To Andrea's point, as I was quoted above is extremely relevant; there's yet to be a good UI for authoring METAFONT programs, and I have many unexplored ideas about this that I developed in the http://metaflop.com and http://metapolator.com projects. I believe with such a gui it could become very power and useful generally. But I have more interesting projects at the moment and for the foreseeable future:)

John Savard · July 2022

Hin-Tak Leung said:

That said, I think one should understand his work in the context that the role of "type designers" didn't/doesn't exist: making a documents/texts visually pleasing is not a very interesting work, and rather tedious, and therefore should be parameterized and automated by mathematics.

Because of the nature of this forum, that statement will be a highly controversial one here.

The people who frequent this forum are people who are type designers, for the most part, and they think of it as interesting and challenging work, and work of an artistic nature that could not, or should not, be done by machine.

And if this is not a statement of your opinion, but rather what you think that Knuth thought, well, I can be fairly confident that Knuth didn't think that either.

However, in fairness, there may be one thing that people should remember before they criticize you.

After all, your name is Leung Hin-Tak. Therefore, while when we think of type design, we think of designing a font with perhaps a few more than 100 glyphs in it, when you think of type design, you may be thinking of designing a font with from 4,000 to 40,000 characters in it, and it's pretty hard for that not to be tedious, however much artistic creativity one may bring to the process.

And in that connection, perhaps you would find Hong-Zi of interest. While the source code is not available on the Web in its original location, it can be found on the Internet Archive's Wayback Machine. The TUGboat article describing it wasn't archived in the Wayback machine, but as it's still in the TUGboat archives, that's not a problem.

It may be in METAFONT 79 code, which doesn't work with today's METAFONT, but if one can get a copy of the old METAFONT to run, this METAFONT Chinese font ought to be usable to generate fonts for certain parameter settings that can then be converted to TrueType.

Of course, the ability to convert to TrueType and the ability to make fonts with Unicode encodings may have been added... at the same time the language changed. There must be some work-around possible.

Hin-Tak Leung · July 2022

Hong-zi seems to be at http://hongzi.sourceforge.net/ , and last updated this side of the millennium, so probably will work with current metafont.

Knuth's work was largely in the 80's, so one should understand his work in that context, before WYSIWYG. Graphic terminals didn't exist, and printers were expensive shared network ones. Computer type design would involve, in those days, writing code on the text terminal to draw without seeing what it would draw, send it to a printer (often in a different room or even a different building), it coming out not quite right, go back to modifying code, send it to another network printer, which is in a different room or perhaps a different building, etc. It is tedious for sure, just walking to a different room to fetch the print-outs. No previewing on computer screen, remember?

DVI was an abstraction based on availability of network printers (and different ones!), and using cheaper or more accessible printers to preview, before sending to the expensive ones. And on-screen dvi previewers, and graphic terminals came later.

To see what curves you have drawn, in Knuth's day, you would need to send an actual print job, from your text terminal, to a shared network printer, located somewhere else, not in your room.

First version of ghostscript, came out in 1989, I think?. No on-screen previewing of postscript on screen before that, either. (At least in the open/academic world).

Hin-Tak Leung · July 2022

It is quite tedious and uninteresting for sure - typing drawing code on a text terminal, having no idea know how it looks like until it comes out at the expensive network printer in the next room/building, perhaps half a day later, look at the print outs.

The printer might jam, ot off-line; go back to text terminal, re-send the print out to a different network printer, finally look at the print out, go back to the text terminal to make some changes, etc.

Repeat this process, and walking back and forth between your text terminal and the printer in the next building, 5 minutes walking/stairs etc away, a few times.

Hin-Tak Leung · July 2022

While we are on DVI, and abstraction of expensive shared network printers, I remember some 80's printing technology. The world now has converged to postscript and PCL, rastering outlines to bitmaps, etc. But in the 80's, there were printers based on having actual pens, programmable with HPGL. Those printers print, by having coloured pens sliding left and right, simultaneously while rollers roll the paper up and down. They are somewhat rare now. When printing is based on having actual pens sliding left right and up down, why shouldn't type design be similar?

John Savard · July 2022

Hin-Tak Leung said:

But in the 80's, there were printers based on having actual pens, programmable with HPGL.

Generally speaking, those devices were referred to as "plotters" rather than printers, and before there was HPGL, there were plotters made by such firms as Calcomp, which were directed to draw lines by means of provided software libraries. However, in those days, computers usually weren't employed in type design; people drew the letters on paper, and then used a pantograph to cut metal, or had their designs photographed to make masters for phototypesetting machines.

Thomas Phinney · July 2022

Hin-Tak Leung said:

Knuth's work was largely in the 80's, so one should understand his work in that context, before WYSIWYG. Graphic terminals didn't exist, and printers were expensive shared network ones. Computer type design would involve, in those days, writing code on the text terminal to draw without seeing what it would draw, send it to a printer (often in a different room or even a different building), it coming out not quite right, go back to modifying code, send it to another network printer, which is in a different room or perhaps a different building, etc. It is tedious for sure, just walking to a different room to fetch the print-outs. No previewing on computer screen, remember?
...
First version of ghostscript, came out in 1989, I think?. No on-screen previewing of postscript on screen before that, either. (At least in the open/academic world).

The first Mac came out in January 1984. Pagemaker in 1985. Windows in November 1985 (although it would not reach great popularity until Windows 3.1 in the 90s). Fontographer in January 1986.

So while the early 80s were certainly command-line based, by the latter half of the 80s things were changing. Even if it did take a little longer to reach the open source world.

Hin-Tak Leung · July 2022

Thanks for the dates for the early GUI-based technology. To quote Wikipedia Metafont entry, the current Metafont is a already a mature(?) re-write, from 1984:

"Donald Knuth started work on font creation software in 1977, and produced the first version of Metafont in 1979. Due to shortcomings in the original Metafont language, Knuth developed an entirely new Metafont system in 1984, and it is this revised system that is used today..."

John Savard · July 2022

Thomas Phinney said:
The first Mac came out in January 1984.

Yes, and it only had 128k of memory, Or, at least, the Superbowl advertisement for the original Macintosh was shown in January 1984. However, since the 512k Fat Mac was launched in September 1984, presumably the first deliveries of the Macintosh had taken place by then.

Ah. The key historical date to be aware of is the release of System 7 for the Macintosh in May, 1991. That was when Apple released TrueType to the public. It was only after that happened that Adobe's Type 1 specification was made available; before then, licensing the right from Adobe to make fonts conforming to it was prohibitive in cost.

December 8, 1992 was when Yggdrasil Linux was first released; if you had 16 megs of RAM in your 386 or 386SX based computer, you could run fvwm on it with reasonable alacrity.

In the other direction, on January 19, 1983, the Apple Lisa became available, though. And Donald Knuth could, no doubt, have afforded one of those.

Basically, then, even with GUI-based machines out there, until May, 1991, Metafont was essentially without any competition... but that didn't seem to help.

Hin-Tak Leung · July 2022

Donald Knuth probably didn't own an Apple machine of any sort. LaTeX wasn't very good until 1995-ish.

I started using LaTeX (and Ghostscript, possibly, by association) between Christmas 1990 and Easter 1991. So I was a Knuth fan boy, before Apple system 7. It was on X11 on Sun OS 4 and Solaris (i.e. Sun OS 5). OSF and CDE came a few years later, so it was probably one of X11 openlook, mwm or fvwm1. I seem to remember being able to switch between those 3 window managers, or at least those widget sets.

More likely, he had a Sun Sparc / DEC / IBM machines in the 198x's (I call that the 90's), or had shared usage of one. By IBM, I don't mean "IBM-compatible PCs", but one of the other sorts...

John Savard · July 2022

Hin-Tak Leung said:

in the 198x's (I call that the 90's),

Why do you do that? It will only confuse everyone, since everyone else means the 199x's by the 90's. I know that the 20th Century goes from 1901 to 2000, so one could indeed say that the 199th Decade goes from 1981 to 1990, but when we say the nineties, we mean the years with "ninety" in them, from nineteen-ninety to nineteen-ninety-nine.

Jens Kutilek · August 2022

I was curious to read more about Knuth’s problems defining a formula to draw the S, and found his article here: https://archive.org/details/dr_dobbs_journal_vol_05_201803/page/n111/mode/2up

Mark Simonson · August 2022

That's also included in Knuth's book TEX and METAFONT: New Directions in Typesetting, 1979. I've got a copy. It's where I first heard of all this in the early '80s. You can still purchase it at places like Amazon, or download a PDF of it here. There was also an expanded version called Digital Typography, 1999.

Howdy, Stranger!

Quick Links

Categories

Metafont – ?

Comments