Keynote Address: The design of TEXandMETAFONT:Aretrospective Nelson H. F. Beebe University of Utah Department of Mathematics, 110 LCB 155 S 1400 E RM 233 Salt Lake City, UT 84112-0090 USA WWW URL: http://www.math.utah.edu/~beebe Telephone: +1 801 581 5254 FAX: +1 801 581 4148 Internet: [email protected], [email protected], [email protected] Abstract This article looks back at the design of TEXandMETAFONT, and analyzes how they were affected by architectures, operating systems, programming languages, and resource limits of the computing world at the time of their creation by a remarkable programmer and human being, Donald E. Knuth. This paper is dedi- cated to him, with deep gratitude for the continued inspiration and learning that I’ve received from his software, his scientific writing, and our occasional personal encounters over the last 25+ years. 1 Introduction 1001 perience handled line breaking, page breaking, and page layout. By the mid 1970s, proprietary compu- 2 Computers and people 1002 ter-based typesetting systems had entered the mar- ket, and in the view of Donald Knuth, had seriously 3 The DEC PDP-10 1002 degraded quality. When the first page proofs of part of the second edition of Volume 2 arrived, he was so 4 Resource limits 1005 disappointed that he wrote [65, p. 5]: 5 Choosing a programming language 1006 I didn’t know what to do. I had spent 15 years writing those books, but if they were going to 6 Switching programming languages 1010 look awful I didn’t want to write any more. How could I be proud of such a product? 7 Switching languages, again 1013 A few months later, he learned of some new devices that used digital techniques to create letter images, 8TX’s progeny 1014 E and the close connection to the 0’s and 1’s of com- 9 METAFONT’s progeny 1014 puter science led him to think about how he himself might design systems to place characters on a page, 10 Wrapping up 1015 and draw the individual characters as a matrix of black and white dots. The sabbatical-year project 11 Bibliography 1015 produced working prototypes of two software pro- grams for that purpose that were described in the 1Introduction book TEXandMETAFONT: New Directions in Typeset- More than a quarter century has elapsed since Don- ting [56]. ald Knuth took his sabbatical year of 1977–78 at The rest is of course history [6] ... the digi- Stanford University to tackle the problem of improv- tal typesetting project lasted about a decade, pro- ing the quality of computer-based typesetting of his duced several more books [61, 65–70], Ph.D. de- famous book series, The Art of Computer Program- grees for Frank Liang [76, 77], John Hobby [33], ming [50–55, 57, 62–64]. Michael Plass [85], Lynn Ruggles [89], and Ignacio When the first volume appeared in 1968, most Zabala Salelles [110], and had spinoffs in the com- typesetting was still done by the hot-lead process, mercial document-formatting industry and in the and expert human typographers with decades of ex- first laser printers. TEX, and the LATEXsystembuilt TUGboat, Volume ?? (2005), No. 0 — Proceedings of the Practical TEX 2005 Conference 1001 Nelson H. F. Beebe on top of it [20–22, 73, 74, 80], became the stan- Computer Science bought one too on the same order. dard markup and typesetting system in the computer Ours ultimately cost about $750 000, and supplied science, mathematics, and physics communities, and many of the computing needs of the College of Sci- have been widely used in many other fields. enceformorethanadozenyears,oftensupporting The purpose of this article is to look back at TEX 50–100 interactive login sessions. Its total physi- and METAFONT and examine how they were shaped cal memory was just over three megabytes, but we by the attitudes and computing environment of the called it three quarters of a megaword. We started time. in 1978 with 400MB of disk storage, and ended in 1990 with 1.8GB for the entire College. Although 2 Computers and people computer time was still a chargeable item, we man- Now that computers are widely available through- aged to recover costs by getting each Department out much of the developed world, and when embed- to contribute a yearly portion of the expenses as a ded systems are counted, are more numerous than flat fee. The operating system’s class scheduler guar- humans, it is probably difficult for younger people to anteed departmental users a share of the machine imagine a world without computers readily at hand. in proportion to their fraction of the budget. Thus, Yet not so long ago, this was not the case. most individual users didn’t worry about computer Until the desktop computers of the 1980s, a charges. ‘computer’ usually meant a large expensive box, at least as long as an automobile, residing in a climate- 3TheDECPDP-10 controlled machine room with raised flooring, and The PDP-10, first released in 1967, ran at least ten fed electricity by power cables as thick as your wrist. or eleven different operating systems: At many universities, these systems had their own BBN TENEX, buildings, or at least entire building floors, called • Compuserve modified 4S72, Computer Centers. The hardware usually cost hun- • DEC TOPS-10 (sometimes humorously called dreds of thousands to millions of dollars (where ac- • cording to the US Consumer Price Index, a million BOTTOMS-10 by TOPS-20 users), and just dollars in 1968 is roughly the same as five million in called the MONITOR before it was trademarked, DEC TOPS-20 (a modified TENEX affection- 2000), and required a full-time professional staff of • managers, systems programmers, and operators. ately called TWENEX by some users), At most computer installations, the costs were MIT ITS (Incompatible Timesharing System), • passed on to users in the form of charges, such as Carnegie-Mellon University (CMU) modified the US$1500 per hour for CPU time and US$0.50 to • TOPS-10, open a file that I suffered with as a graduate student On-Line Systems’ OLS-10, earning US$1.50 per hour. At my site, there weren’t • Stanford WAITS (Westcoast Alternative to any disk-storage charges, because it was forbidden • ITS), to store files on disk: they had to reside either on Tymshare A UGUST (a modified TENEX), punched cards, or on reels of magnetic tape. A cou- • ple of years ago, I came across a bill from the early Tymshare TYMCOM-X, and on the smaller • 1980s for a 200MB disk: the device was the size DECSYSTEM 20/20 model, TYMCOM-XX. of a washing machine, and cost US$15 000. Today, Although the operating systems differed, it was usu- that amount of storage is about fifty thousand times ally possible to move source-code programs among cheaper, and disk-storage costs are likely to continue them with few if any changes, and some binaries to drop. compiled on TOPS-10 in 1975 still run just fine on I have cited these costs to show that, until desk- TOPS-20 three decades later (see Section 3). top computers became widespread, it was people Our machines at Utah both used TOPS-20, but who worked for computers, not the reverse. When Donald Knuth’s work on TEXandMETAFONT was a two-hour run cost as much as your year’s salary, done on WAITS. That system was a research op- you had to spend a lot of time thinking about your erating system, with frequent changes that resulted programs, instead of just running them to see if they in bugs, causing many crashes and much downtime. worked. Don told me earlier this year that the O/S was aptly When I came to Utah in 1978, the College of Sci- named, since he wrote much of the draft of the ence that I joined had just purchased a DECSYSTEM TEXbook while he was waiting in the Computer Cen- 20, a medium-sized timesharing computer based on ter for WAITS to come back up. By contrast, apart the DEC PDP-10 processor, and the Department of from hardware-maintenance sessions in a four-hour 1002 TUGboat, Volume ?? (2005), No. 0 — Proceedings of the Practical TEX 2005 Conference Keynote Address: The design of TEXandMETAFONT:Aretrospective block each week, the Utah TOPS-20 systems were Mark Crispin’s mail client, mm,stilloneofthe rarely down. • best around [Stanford]; For about a decade, PDP-10 computers formed Will Crowther’s adventure, Don Daglow’s base- the backbone of the Arpanet, which began with • ball and dungeon, Walter Bright’s empire,and just five nodes, at the University of California cam- University of Utah student Nolan Bushnell’s puses at Berkeley, Los Angeles, and Santa Barbara, pong, all developed on PDP-10s, were some of plus SRI (Stanford Research Institute) and Utah, the earliest computer games [Bushnell went on and later evolved into the world-wide Internet [24, to found Atari, Inc., and computer games are p. 48]. PDP-10 machines were adopted by major now a multi-billion-dollar world-wide business computer-science departments, and hosted or con- driving the computer-chip industry to ever- tributed to many important developments, including higher performance]; at least these: part of the 1982 DISNEY science-fiction film • Bob Metcalf’s Ethernet [Xerox PARC, Intel, and TRON was rendered on a PDP-10 clone [cu- • DEC]; riously, that architecture has a TRON instruc- tion (Test Right-halfword Ones and skip if Not Vinton Cerf’s and Robert Kahn’s invention of the masked) with the numeric operation code 666, • Transmission Control Protocol and the Internet leading some to suggest a connection with the Protocol (TCP/IP); name of the film, or the significance of that the MACSYMA [MIT], REDUCE [Utah] and • number in the occult]; MAPLE [Waterloo] symbolic-algebra languages; Frank da Cruz’s transport- and platform-inde- several