BijlageN KnuthmeetsNTGmembers 38 Knuth meets NTG members March 13th, 1996 Abstract OnJanuary6th1996,KeesvanderLaaninformedtheNTGthatDonaldKnuthwouldbeinHollandinMarch. KnuthwasinvitedbytheMathematischCentrum(MC,nowadayscalledCentrumvoorWiskundeenInfor- matica,CWI)becauseofCWI’s50thanniversary. BothKnuthandMandelbrotwereinvitedasspeakersat thecelebration. TheNTGnoticedthatthiswasanexceptionaloccasiontoorganizeaspecialmeetingwithKnuthforallDutch TEXandMETAFONTuserswhowouldliketomeettheGrandWizardhimself. FortunatelyKnuthacceptedtheNTGinvitationandsoameetingwasorganizedin‘DeRodeHoed’inAm- sterdamonMarch13th. About35peoplefromalloverthecountryandevenfromBelgiumjoinedtomeet Knuth. EverythingwasrecordedonbothvideoandaudiotapebyGerardvanNes. ChristinaThielevolunteeredto writethistranscript. ErikFrambach: Welcome,everyone. Thisisaveryspe- questions-and-answersto try to keep me honest, because cialmeetingon theoccasionthatMr. DonaldKnuthisin theyalsodidthatinPrague. Soincasethesamequestion Holland. TheNTGthoughtitwouldbeagoodideatotake comesup, you’llhaveto take the averageof the two an- theopportunityandaskhimifhewouldbewillingtoan- swers.[laughter] swerourquestionsaboutTEX,METAFONT,andanything WietseDol: DidyouknowthatBarbaraBeetondoesthat? elseconnectedtothethingswedowithTEX. Luckily,he Shemailsyouandsays“Tapeeverything.” hasagreed. SoweareveryhappytowelcomeMr.Donald Knuthhere—thankyouforcoming. Knuth: Yes,that’swhattheysaidinPraguetoo![laugh- Tonightwehavetimetoaskhimanyquestionsthatwehave ter]Ithinkshe’sdesperateforthingstodo,ormaybeshe longbeenwaitingtoposetohim[laughter]. I’msurethat justhasalotofquestions. ButbeforeIopenquestions,let allofyouhavemany,manyquestionsthatyouwouldlike mesaythatoneofthemostinterestingquestionsaskedme theGrandWizard’sopinionabout. So,wecouldstartnow in Praguewas after the session. And I wish it would get withquestions. into[therecord].Thequestionwas: howdidImeetDuane Bibby,whodidtheillustrationsforTheTEXbookandThe METAFONTbook? Ialwayswantedpeopletoknowabout thatsomehow. Here’s the story. I had the idea that after writing math books for many years, I wanted to have a book that had moreweird—well,anyway,different—illustrationsinit. HereI waswritinga bookaboutbooks—bookshaveil- lustrations,sowhyshouldn’tIhaveillustrationstoo. So,I wrotetoanartistcalledEdwardGorey.Doesanyoneknow ... FransGoddijn: Yup.Amphigorey. Beautiful. Knuth: Yes, Edward Gorey. Amphigorey. He makes very morbid drawings but with a wonderfulsense of hu- mor. I hadusedseveralofhisbookswith mychildren. I DonaldE. Knuth: I getto ask questionstoo! [laughter] thoughthewouldbeanaturalperson. Iwrotehimtwolet- LastSaturdayIwasinPragueandtheCzechoslovakTEX ters but he never responded. Then I wrote to a Japanese usershadasessionsomethinglikethisandyou’llbeglad artistcalledAnno,MatsumasaAnno,whoisreallythelogi- toknowthatIsawquiteafewcopiesof4TEXCD-ROMs calsuccessortoEscher. [...]AnodoeswhatEscherdidbut atthatmeeting. incolor,soIaskedhimifhewoulddoit. Hesentmeback aniceletter,saying“I’msorryIdon’thavetimebecauseI It’snotmyfirsttimeinAmsterdam.IwasinAmsterdamin havesomanyothercommitments,butherearefiveofmy 1961,soit’sonlybeen35years,andprobablylessthan35 yearstillthenexttime. Iguessthey’retaperecordingthese BijlageN KnuthmeetsNTGmembers 39 booksfullofpicturesandifyouwanttouseanyofthose, takethetimetodoagood,carefuljobwithit, thenaswe goahead.” Iwantedpersonalizedpictures. learnmoreabouttypesetting,itwillhappenthatsomething elsewillcomealong.IpersonallyhopethatIwon’thaveto ThenIwenttoapartyatStanfordwheretherewasalady taketimetolearnanewsystem,becauseIhaveenoughfor who worked for a publisher. She’d just met a brilliant myownneeds. ButIcertainlyneverintendedthatmysys- youngartistwhoshe’djustworkedwith. Iinvitedhimto temwouldbetheonlytoolthatanybodywouldeverneed cometo my house, and we spentsometimetogetherand fortypesetting. ItriedtomakeitasgeneralasIcouldwith he’sawonderfulperson. Duanelivesnowupinnorthern areasonablysmallprogram,andwithwhatweknewand California,about4hours’drivefrommyhouse,soIonly understood about typesetting at the time. So these other wentupthereoncetoseehim. Hesometimescomesdown projects— I don’tconsiderthatthey’rea threatto meor totheSanFranciscoareaonbusiness. Firstwediscussed anything. I hopethattherewillbe somecompatibilityso thebookandthenhesentmeabunchofdrawingsandall that—Imean,I’dliketobeimmortal—sothatthebooks kindsof sketchesthathe had. Originally,TEXwas going I’vewrittennowcouldstillbetypeset50yearsfromnow tobeaRoman,andhedrewthismaninatogawitholive withouthavingtogothroughthefilesandeditstuff. Ilike branchesinhishead—whichiswhythelionhastheolive thearchivaland machine-independentaspects of TEX es- branchesnow.Butallofasuddenhestarteddoingsketches pecially,andItriedtosetamodel,aminimumstandardof of his cat, which really seemed to click, and pretty soon excellenceforotherpeopletofollow. hehadadraftofall35orwhateverdrawings,usingalion. Mostofthoseeventuallybecomethedrawingsinthebook, HansHagen: butwhenyoulookinthefuture,...youcon- andweadjustedhalfadozenoftheothers. WhenIwentto sidertoday’sprogrammingbyalotofpeopleasanart,well visitupathishouse,I gottomeetTEXthecat. Helooks alotofarttakeshundredofyearstoberecognizedasart. In verymuchliketheoneyouseeinthebook. Sothat’sthe aboutahundredyearstherewillbeprettydifferentcomput- storyaboutDuaneBibby. ers,theprogramminglanguageswillbechanged,themedia onwhichweputallthosethingswillbechanged.Realpro- Erik: Thank you. Who would like to start with the first gramsandeverythingrelatedtothem,willtheyeverhave question? Pleaseidentifyyourselfwhenyouaskone. achancetobecomeimmortal,asyouseeit? Piet vanOostrum: Mynameis Pietvan Oostrum. You Knuth: Didyoustateyourname?[laughter] havethiswonderfullionontheTEXbookandthelionness ontheTheMETAFONTbook. Whataboutbabylions? HansHagen: I’mHansHagen. Knuth: Oh, I see ... [laughter]. Duane still does illus- Knuth: You’resayingthatit’sprettyarrogantofustoas- trationsforspecialoccasions. He’smadeillustrationsfor sumethatwhatwedonowwilllastat all. Technologyis theJapanesetranslationofboththeTEXbookandtheThe changingsofastthatwehaveabsolutelynoideawhatpeo- METAFONTbook. HehasTEXandMETAbothdressedup ple are going to think of next. One hundred years ago, inJapanesecostumes. Sonow,iftherehappenstobesome physicists were saying there was nothing more to do in kind of an offspring that would come out of some other physics, except to get another decimal — a fifth decimal user, then, I imagine he would glad to help do it. But it placeforthefundamentalconstants—andthenthatwould wouldprobablybealittlebitofanillegitimatechild,from wrapupphysics. So,thereisnowaytoknowaboutthese my point of view [laughter]. I mean, I wouldn’ttake re- things. ButIdobelievethatoncewehavethingsinelec- sponsibilityforanythingthosecharactersdo[laughter]. tronicformandwehavemirrorsitesofthem,thereisafair degreeofimmortality—whereaspaperburns. Do you know anything about this project called ‘The Clock’, being developed by Stewart Brand and his col- leagues?He’stheonewhopublishedtheWholeEarthCat- alogue.Theyhaveabunchofpeoplethatareconsideringif theycouldbuildsomethingthatwouldlastforathousand years... Idon’twanttogoontoomuchmoreaboutthat. IdohopethatthestabilityofTEXwillmakeitpossibleto reproducethethingswe’redoingnow,later. Andsinceit’s fairlyeasytodothat,Ithinkitwillhappen—unlessthere’s a nuclear holocaust. Some mathematicians have this de- bateaboutthePlatonicview... doeseverythingin math- ematicsexistandwe’rejustdiscoveringit,orareweactu- allycreatingmathematics? Insomesense,oncesomething Piet: So what are your ideas aboutthe offspringof TEX andMETAFONT? gets put into bits, it’s mathematics and thereforeit exists forever,evenifthehumanracediesout—it’sthere,butso Knuth: Well,Ithinkthatnomatterwhatsystemyouhave, what? there will be a way to improveit. If somebodywants to Erik: Who’snext? BijlageN KnuthmeetsNTGmembers 40 MarcvanLeeuwen: IfIcouldextendabitontheprevi- now,togettheArabicnames,IhavetouseArabTEX,toget ousquestions. ThestabilityofTEXitself,Icouldimagine, theHebrewnames...IhadaterribletimetryingtogetHe- mightbeastumblingblockfordevelopmentofnewthings brewfontsonCTANtwoweeksago—Icantellyouthat exactlybecauseit’ssostableandeverybody’salreadyus- wholestoryifyouwanttoknow... Ikeptclickingonthe ingit. Soifsomethingcomesalongthatisjustabitbetter, differentthingsandtheywouldrefertofilesthatdidn’tex- thenpeoplewillnottendtousethatbecauseit’snotavail- istandREADMEfilesthatwerefouryearsoutofdateand ableeverywhere,andthereareallkindsofreasonstokeep inconsistent,soIcouldn’tfindanyHebrewfonts. Maybe onusingtheoldthing. youhaveitonyourCD... Knuth: IguessIsaidinFloridathatpeoplearestilltrying Johannes: I could certainly point you to someone who tousetheoldfontsthatI’mstilltryingtostampoutfrom couldhelpyouwiththeHebrewfont—Iknowsomeone theworld.FouryearsagoIredesignedtheGreeklowercase inIsraelwho’stryingtodoHebrewsupportwithintheBa- delta and I made the arrowheadsdarker. I didn’t change belsystem. AndtheydodotypesettinginIsraelwithTEX. anythinginthewayTEXoperates—allthedimensionsand Knuth: MyowntypesettingfriendinIsraelisDanBerry, the characters’ heights and widths stay exactly the same. who unfortunatelyis fairly committed to troff [laughter]. ButIdidtuneupalotofthecharacters. StillIseelotsof I’m sure that I can get good Hebrew throughYannis and math journalsarestill using theold onesfromfouryears ago,andIgetlettersandpreprintsfrompeoplewiththeold- Omega. I sure hope UNICODE is going to arrive sooner ratherthanlater; it’smuchbetterthanthealternativesfor styledelta. IchangeditbecauseIjustcouldn’tstandtheold muchthereasonsthatMarc [vanLeeuwen]mentioned. I versions[laughter]. NowI’vegothomepages—ifIever have some errata to TEX or something I put them there: haven’tfounda great enthusiasm in Japan for UNICODE, becausetheyhaveasystemthatseemstoworkprettywell http://www-cs-faculty.stanford.edu/˜knuth. forthem,so whychange. EverytimeIask aJapanesefor This gets to my home page, and there’s a reference say- ing,‘ImportantnoticeforallusersofTEX’,andthatpage hisnameinUNICODE,he’llsay,‘what’sUNICODE? Here’s says‘Lookatthelowercasedeltaandifyouhavethewrong myJISname’. ButtheJIScharactersdon’tincludetheChi- nesecodes,andinfact,myownname—IhaveaChinese one,youdie!’[laughter] name— and my namein JIS isn’tquitethe same. There Iunderstandthatpeoplehaveareluctancetochangefrom aretwodifferentUNICODEcharacters,onefortheJapanese somethingthatthey’vebecomeaccustomedto. Iknowof versionandonefortheChinese. two main successors to TEX: one is "-TEX and the other isNTS."-TEXisgoingtobeapparently100%compatible Intheback? Kees? with TEX, so if somebody doesn’t switch over to incom- KeesvanderLaan: I havealotofquestionsofcourse. patible features, then they have a system that still works ButIwouldliketostartwithsomequestionsaboutMETA- witholdthings.Thatwillallowagradualchange-over.It’ll FONT. The first oneis: how comemacrowriting in TEX takemorespaceonacomputer,ofcourse,butthat’snota andMETAFONTissodifferent? big deal these days. The people who work on "-TEX al- wayssentmeveryreliablecommentsaboutTEXwhenthey Knuth: WhyaremacrosinTEXandMETAFONTsodif- caughterrorsinmystuff,soIimaginethey’regoingtobe ferent? I didn’t dare make TEX as extreme as META- doinga carefuljob. So it’ll beoneof thesethingswhere FONT.Theselanguagesareofcompletelydifferentdesign. youwalk into a randominstallation of UNIX or whatever METAFONT is insome waysanincredibleprogramming andyou’llfind"-TeXthereasthedefault,andyou’llstill language—it’sobject-orientedmacros. Youhavemacros haveTEX. Then youalso havecertain otherfeaturesthat inthemiddleofrecordstructures. mightbereallyimportanttoyouforyourspecialapplica- ThewayIdesignedtheselanguagesisfairlysimpletode- tions. scribe. Let’s take TEX. I wrote down one night what I JohannesBraams: Youmentioned"-TEXandNTS.But thoughtwouldbea goodsourcefileforTheArtofCom- areyoualsoawareoftheOmegaproject? puterProgramming.ItookalookatVol.2,whichIhadto typeset. I started outon the first page, andwhen I gotto Knuth: Oh, theOmegaproject? Yes, I’mhopingto use any copythat lookedvery muchlike somethingI had al- thatmyselffortheauthors’namesinmyTheArtofCom- ready done I skipped that. Finally I had examples of all puterProgramming.I’vebeencollectingthenamesofChi- thedifferentkindsoftypesettingconventionsthatoccurin nese, Japanese, Indian, Hebrew, Greek, Russian, Arabic Vol.2. Ittotalled5printedpages—andyoucanevensee authorsandIwanttotypesettheirnamesproperly[laugh- thesepages—exactlywhatmyoriginaltestprogramwas ter], not just in transliteration. I have some rudimentary —inapaperbyDavidFuchsandmyself,wherewetalked softwarethatwilldothisforproofingpurposes,forgetting aboutoptimumfontcaching.1 Inthere,wegaveanexam- mydatabasegoingandforwritingtopeopleandsaying,‘Is pleandweshowthese5pages,whichwouldillustratewhat thisyourname?’ WiththeOmegasystem,I’mhopingthat IwantedTEXtobeabletodo. IwroteoutwhatIthought it’llbeaccompaniedbygoodfontsthatwillmakeitpossi- Iwouldliketotype—howmyelectronicfileshouldlook. bleformetodothiswithoutawholepileofwork. Right Andthen,Isaid,OK,that’smyinput,andhere’smyoutput 1ACMTransactionsonProgrammingLanguagesandSystems7(1985),74. BijlageN KnuthmeetsNTGmembers 41 —howdoIgetfrominputtooutput? Andforthis,well,it system that goes beyondthebasics. I always wanted the lookslikeIneedmacros[laughter]. systems that I would make widely available would be be abletohandle99%ofallapplicationsthatIknew. ButIal- SamethingforMETAFONT.Iwentthroughmyfirstdraft waysfeltthereweregoingtobespecialapplicationswhere of all the fonts that later became Computer Modern. I theeasiestthingwouldbetochangetheprogram,andnot wrote actually in SAIL, an Algol-like compiler language, writeamacro. butSAIL hadamacroability,soIdevelopedafewprimi- tivemacrosinwhichIcouldsay,‘pickupthepen’,‘draw Itriedtomaketheprogramssothattheywouldhavelog- frompoint1topoint2’,andthingslikethat. Thesemacros ical structure and it would be easy to throw in new fea- were compiled by the SAIL compiler into machine lan- tures. This hasn’t happened anywhere near as often as I guage, whichwouldthen drawthe letters. I wentthough thought because people were more interested, I think, in theentirealphabet,andbytheendoftheyear,Ihadsome inter-changeability of what they do; once you have your 300littleprograms,eachonedrawingaletter. ThenIre- own program, then other people don’t have it. Still, if I alizedwhatkindofalanguageIwouldwanttowritein,to werealargepublisher,andIweretogetspecialprojects— describetheletters. Sooneday,onafamilycampingtrip someencyclopaedia,someneweditionoftheBible,things — I was in the Grand Canyonwith my wife and kids — likethat— I wouldcertainly thinkthatthe rightthing to Itookanhouroff,satunderatreeandwroteoutthepro- dowouldbeto hireagoodprogrammerand makeaspe- gramfortheletterA,inalanguagethatIthoughtwouldbe cial computer system just for this project. At least, that agoodalgebraiclanguage,reflectingatahighlevelwhatI wasmy ideaaboutthe way peoplewould doit. It seems hadbeendoingwithprettyprimitivelow-levelinstructions thathasn’thappenedverymuch,althoughinBrnoImeta inmySAILprograms.IdidtheletterB,too. CapitalAand studentwhoiswellalongonproducingAcrobatformatdi- B,andthenwentbacktothecampingtrip. Thesesheetsof rectlyinTEX,bychangingthecode. AndtheOmegasys- paperwhereIhavemyoriginalprogramsarenowinStan- temthatyoumentioned,that’s150,000linesofchangefiles ford’sarchive—theprogramfortheletterBwaspublished [laughter]. IbuiltinhookssothateverytimeTEXoutputs in a Stanford library publication calledImprintlast year. apage,itcouldcometoawhatsitnodeandawhatsitnode Thewomanwho’sinchargeofrarebooksandmanuscript couldbesomethingthatwascompletelydifferentin each collectionsatStanfordisquiteinterestedinMETAFONTso versionofTEX. So,whentheprogramseesawhatsitnode, shewrotealittlearticleaboutwhattheyhave. itcallsaspecialroutinesaying,‘howdoItypesetthiswhat- sitnode?’ It’lllookatthesub-typeandthesub-typemight ThatprogramagainimpliedthatIwantedsomemacrosto beanothersub-typeputinasademooritmightbeabrand- go with it. But these neededto be muchmorestructured newsub-type. than themacrosof TEX. It hadto bethat whenI said, ‘z 1prime’,thiswouldactuallybeequivalentto‘(x1prime, Similarhooksarein theMETAFONT program. Ifpeople y 1 prime)’ and I wanted to be able to write, ‘z 1 prime’ haveextratimewhenthey’renotbrowsingtheWeb[laugh- withoutanydelimiters. Itturnedoutthatinordertohave ter],Irecommendasagreatrecreationtoreadtheprogram a high-levellanguagethat would feel naturalto me writ- for METAFONT. Some parts of it are pretty rough going ingtheprogram,it hadto lookcompletelydifferentfrom andIhopethatnobodyeverfindsabugtherebecauseI’d TEX. SoTEXandMETAFONTshareacommonformatfor hatetohaveto lookatthem[laughter]. Butthosearethe errormessagesandcertainotherdatastructuresinside,but rasterizationroutines,thethingsthatactuallyfillinthepix- otherwise,they’requitedifferentsystemsbecause,inorder els. Therearemanyotherthingsinthatprogram—thelin- to have a good high-levellanguage, I don’twant to have earequationsolverthatithasandthedatastructureabilities towastetimewritingparentheses,brackets,commas,and ...lotsofbeautifulalgorithmsareinthere—totakesquare otherdelimiters. rootsinfixedpoint,andtheintersectionoftwocurves,and soon. METAFONTisfulloflittleprogramsthatweregreat Kees: It’s a nice introduction to my second question fun to write and that I think are useful and interesting in [laughter]:ForthefutureofMetaPost,whichallowsmark- theirownright. IthinkwhenJohnHobbywroteMetaPost, upofpictures,with.epsastheresult,whatisyouratti- heenjoyedit,becausehecouldaddhisownnicelittlepro- tudeto2.5dforMetaPostandMETAFONT?Forexample, gramstotheonesthatarealreadythere. addingatripleasananalogyofthepaireddatastructure? I’mabigfanofMetaPostfortechnicalillustrations.Idon’t Knuth: MetaPost alreadyhas a datastructurefortriples knowanythingthat’snearasgood,soI’mdoingalltheil- becauseofcolor. SoRGBareactuallytriplesofnumbers. lustrationsofTheArtofComputerProgramminginMeta- Post. Also, the technicalpapersI’vewrittenaregoingto Kees: Yes,butthetripleasadatapointinspace? be published in a series of eight volumes by Cambridge Knuth: Ah,Isee. IdidwriteMETAFONTinawaythat UniversityPress, andalltheillustrations, exceptthepho- tographs,aregoingtobeMetaPosted. Thefirstvolumeof hashooksinitsothatitcanbeeasilyextended;[forexam- theseeightwasthebook,LiterateProgramming;thesec- ple], ifyouwantto draw3-dimensionalpictures,forper- ondvolumeis going to come outthis summer and is go- spectiveandprojectivegeometryinsteadofaffinegeome- try. TheprogramitselfforMETAFONTwaswrittensothat ingto becalledSelectedPapersinComputerScience. It reprintsadozenor15papersthatIwroteforgeneralaudi- itcouldeasilybechangedbypeoplewhowantedtohavea BijlageN KnuthmeetsNTGmembers 42 ences, notforspecialists in computerscience, butinSci- ingthecomputerinterpolate. Themultiplemasterfontsin entificAmericanorSciencemagazineandthingslikethat. PostScript allow up to four parameters, and almost all of Thethirdvolumewillbeaboutdigitaltypography,andit’ll themhaveonlyoneortwoparameters. ThemostIknow reprintallmy articlesinTUGboatand thingsaboutTEX. of is two; probablyothers have goneall the way to four. Whatdoyouthink,bytheway—shouldIpublishinthat Butthentheyhavetoprovidedrawingsforalltheextreme thirdvolumethememothatIwrotetomyselfthefirstnight, pointsoftheseparameters. whenIdesignedTEX? Iputitinacomputerfileandit’sin In spite of this limited use of parameters, what’s avail- thearchives,butI’venevershownittoanyone. [roundof ablecommerciallyis quite beautiful, as far as readability “ofcourse!” and“sure”andlaughterfromtheaudience]. isconcerned,althoughitdoesn’treallyprovidethequality Maybeit’dsellmorebooks[morelaughter]. thatyou guyshad in the Netherlandsin the 17thcentury. FransGoddijn: Youneedtoputitonyourhomepageand What’stheman’sname,thegreatpunchcutteratEnschede´ wecanthendecide— — he made 4.5, 5pt up to 16pt, and each letter was de- signedforitssize,andfontshadaniceuniformappearance. Knuth: No, no. That way we’d never sell the books Thiswouldn’thavehappenedatallwith theType1fonts. [laughter]. Not that I’m a mercenary type of person, of There were two guys who did most of the punch cutting course. It’sinafilecalledTEX—well‘teks’, actually. I forEnschede´andothersinthe18thcentury: Oneofthem, havetoadmitIpronouncedit‘teks’foramonthortwo— Fleischman, was a genius for really beautiful letters; the Iwasthinkingof‘technicaltexts’,though.tex.onewas other,Rosart,wasjustgoodatmakinglotsandlotsoflet- thenameofthefileanditwouldmakeinterestingreading ters[laughter].2 [...] Theywerefun. Rosartwouldmake probably,someday. allkindsofhighlydecoratedalphabetsandthingslikethat. I have a big coffee-tablebook that gives examples of all Andyournameis? thefontsfromEnschede´,whichwastranslatedintoEnglish JanKardan: InthiscompanyIwillprobablyaskavery byMatthewCarter’sfather. Anyways,inthisbook,Type- hereticquestion, but a little heresy makesa lot of fun — foundriesintheNetherlands, you can look atthese type- talkingaboutMETAFONT.Thereareprobablymanytype facesandweep.3 foundries now [that] crank out lots of good-qualityfonts Still,onalaserprinter,wegetprettygoodfontsnow,and andkerningtables.It’snotclearwhetherPostScriptorTrue therefore it looks like there won’t be that many profes- Typewillsurvive. DoyouthinkthatMETAFONTwillsur- sional type designers using METAFONT. Pandora was a vivetextfonts? Nottalkingaboutthemathfonts. gooddesignbyagenuinegraphicartist. METAFONThas Knuth: Idon’tthinktheextracapabilitiesofMETAFONT turned out to be wonderful for making ordered designs andspecial-purposethingsforgeometry.There’snowthis haveprovedtobenecessaryforgood-qualitytypefonts,al- really neat system in Poland where they have TEX and though I think that you can still make better-quality type METAFONT in a closed loop — TEX outputs something fontswithit. Designersfinditdifficulttothinkasacom- andthenMETAFONT drawsacharacterandif thatdoes- puter person does, in the sense that when people in the n’tfit,TEXsays,‘gobackandtryitagain’. Jackowskiand computerbusinessautomatesomething,tryingtomakethe Ryc´kounderstandTEXandMETAFONT,andtheprograms computerdosomething,it’snaturalforustohaveparam- arewelldocumentedandcandothesethings. SoMetafont eters and say that we’re going to try to solve more than isn’tgoingtodisappearforthatreason;butit’snevergoing oneproblem. Wetrytosolveawholevarietyofproblems tobetaughtinhighschool. basedontheparametersthatpeopleset. Butit’smucheas- ier if people gave us only a single problem with a single parameter, then we could have the computers do exactly theprescribedthing. Computerscientistshavebecomeac- customedtothinkingofhowwewouldchangebehavioras conditionschange, butdesignersaren’tat all accustomed tothis. Theyaremuchhappierifthebosssaysonemonth, “Givemearomanfont,”andthenextmonth,“Givemea boldfont.”It’smuchmoredifficulttosay,“Showmehow youwoulddrawsomethingnomatterhowheavyIwantthe letters to be.” METAFONT provides a way to solve that problemandtodrawcharacterswithparameters,butit’sa raredesignerwho’scomfortablewiththatnotion.Theycan domultiplemasterfontsbymakingmultipledrawingsand then matching up points between the drawings and hav- 2JohannMichaelFleischman,1701–1768;Jacques-Fran¸coisRosart,1714–1777. 3TypefoundriesintheNetherlandsfromtheFifteenthtotheNineteenthCenturies,byCharlesEnschede´,translatedbyHarryCarter (Haarlem: StichtingMuseumEnschede´, 1978), 477pp. Thismagnificentbookwascomposedbyhandandprintedbyletterpressto commemoratethe275thanniversaryofJoh.Enschede´enZonen. BijlageN KnuthmeetsNTGmembers 43 FransGoddijn: MynameisFransGoddijnandIhaveone makeitmoregeneralandsoon. ButIdidwantatoolfor questionwithsomesub-questions[laughter];I’dliketoask myselfby whichI couldproducebooksthatwouldmake the sub-questions first. What I’m wondering — and this mefeelgoodafterspendingalmostallmylifewritingthose mayhavebeenaskedoftenbefore—iswhetheryouwould books. consider,inretrospect,whatyouhavecreated[tobe]anart IstartedwritingTheArtofComputerProgrammingwhen oratool? AndthereasonIaskis—whenIhearyouspeak Iwas24yearsoldandIstillhave20yearsofworktodoon withsomuchpassionfortypefontsandthebeautifulalgo- it. That’salotoftime. Idon’twanttowritethosebooksif rithmsthatyouputintoMETAFONTthatyouwouldliketo they’regoingtocomeoutlookingawful. Iwantedaway pointpeopletoandtherecognitionthatyougetfrompeo- tomakeitpossible[toproducegood-lookingbooks].Orig- plewhounderstandthat—but,thereisavastmajorityof inally, whencomputersstarted out, they knewonlynum- users who just got TEX from some server, never realized bers,digits. The19th-centurycomputerscouldprinttables. whocreatedit,anduseittotypesetnotalwaysverypretty Thenwehadcomputersthatcoulddonumbersandletters, documents[laughter].Theydothatinaverycrudewayand butonlyonateletypemachine;soyouhadsomecapitallet- don’tcareless. YoufrozeTEXatacertainpoint,allowing tersanda32-characterset. Butthen,afterIgraduatedfrom otherpeopletobuildaroundit. Iwaswonderinghowsuch college,wegot... letmesee,Iwasprobablytenyearsout athingwouldfeeltoafather—areyoufatherofapiece ofcollegebeforewecoulddolowercaselettersonacom- ofartthatotherpeopleuseasatool,orisitachildthatyou puter. Youknow,thePASCALlanguage,whenitcameout, havefrozeninitsdevelopment,thatwillnevergrowup... it used all uppercase letters — there was never any con- therearesomanyquestions...ifyoujustgobacktotheart siderationthat therewouldbemorethan 64charactersin vs.toolidea,andyourfeelingsaboutthat. acomputer’srepertoire. Finally,wewerebeginningtosee Knuth: Obviously, if I write somethingthat hasa lot of inthemiddle70sthatcomputerscouldactuallydolower- powertodomanydifferentthings,it’llbepossibletomake caseletters,andproducesomethingthatlookedalittlebit it do awful things. I just came from the Rijksmuseum, readable,alittlebitlikebooks. Wow![laughter]. wheretheyhaveanexhibitcalled“TheAgeofUgliness”. Thentherewasthisdevelopmentoftypographicsoftware It was a wholebunchof fancy silver bowls fromthe late starting at MIT in 1960 and going through 4 or 5 gener- 19thcentury.... Whenyousayanart, I’mnotsureIun- ations, leading to troff and EQN, where there was even derstandexactlywhatyoumean. Tome,artisusedintwo mathematics being typeset. In 1977 I therefore knew an quitedifferentsenses,mostoftennowadaysinthesenseof existence theorem: It was possible to typeset something fineart,whileart,originally,Kunst,wasanythingthatwas thatlookedalmostlikegoodmathematics. EQNwasbeing not natural — so we have the word artificial, something usedinphysicsjournalsandexperienceshowedthatsecre- that is made by people instead of by nature. The Greek tariescouldlearnhowtodoit. SoIthought,“Whynotgo wordis techne[laughter]. Butthen youreferto atoolas all the way to the end, to convergence?” What I wanted somethingthatismaybejustadevicethatisthefastestway todowithTEXwasnottobealittlerefinementovertroff togetfromheretotherebutmaybeyoudon’tcareaboutel- and the other things, but I was saying now, “Let me try egance... ButwhatIthinkyoumeanwhenyoutalkabout togotothebesttypographythat’severbeenachievedby artistheaesthetics—somethingaboutbeautyandsome- mankind”Exceptfortheilluminatedgoldleaftypeoflet- thingwithalittlebitofloveinit. WithTEX,myideawas tering, I wanted to at least — when it came to black and tomakeitpossibletoproduceworksthatyouareproudof; whiteprinting—I wantedto matchthe bestconventions Iassumedthatpeoplecanenjoyactuallyspendingalittle that had been achieved. Computer typesetting had gone extratimemakingtheresultsbetter. Ididn’texpectthatthe throughthislengthydevelopment,gettingalittlebetterand wholeworldwouldbedoingthis[laughter]. alittlebetter. Itwastimetosay,“Well,let’sjumptotheend Incidentally, I can’t understandthe mentality of a person now.” Of course, I didn’tthink this would be an activity who writes graffiti on a beautifulbuildingalthoughI can thateverybodywouldwanttodo. Buttherewereenough see why drawing is fun. Why would you want to scrawl peoplethat would care abouttrying to get as much qual- something — some kind of animal instinct of territory ityaspossible,thattheycouldbe—well,that’swhyIfi- mightaccountfor it, I suppose, but it’s really impossible nallymadeTEXmoreavailable. TheAmericanMathSo- formetoconceiveofsuchactions. cietywerethefirstpeople,nearlythefirstpeoplewhocon- vinced me that I should make the system do more than I When it comes to matters of aesthetics, you can’t dictate originallyintended. taste. You can’t say that your idea of beauty is going to matchanyoneelse’sideaofbeauty. ButIdidwanttohave AndriesLenstra: Whydidn’tyoustartfromtroff? Itwas a tool where we could reach the highest levels of beauty completelyinappropriate? accordingtoourowntastes. Ididn’tallowpeopletohave Knuth: Yes,yes. Yousee,troff waspatchedontopof... letterspacingveryeasily,butItriedtomakeeverythingelse Imean,therewasawholesystem,itwasafifthgeneration, easy[laughter]. .... Ofcourse,IoriginallydesignedTEX eachofwhichwasapatchonanotherone. Soitwastime justformyself, forTheArtofComputerProgramming;I to scrap it and start all over again: “Here’s what the lan- thoughtmysecretaryandIweregoingtobetheonlyusers. guageshouldbe,solet’sdesignsomegooddatastructures Anditwasn’tuntillaterthatIwasconvincedthatIshould BijlageN KnuthmeetsNTGmembers 44 forit.” Not“Let’strytobecompatible.” Ihadtheadvan- wholefieldofartificialintelligencehasbeenoneofthear- tagethatIwasnotatBellLabs,whereIwouldn’tbehurt- easthathashadgreatestspin-offstocomputersciencebe- inganybody’sfeelingsbysaying,“Let’sthrowitallaway” causethey’vetriedtosolveveryhardproblems.Especially [laughter].ItwasimpossibleforthepeopleatBellLabsto in theearly days, they came up with methodsthat turned dosuchathing—itwouldn’tbenice. Butitoccurredto outtobeusefulinmanyotherpartsofcomputerscience. methatnowthatwehadproofthatthisgoalwaspossible,I So,it’smyfeelingthatwhenpeopleareworkingonmore shouldstartover,andrethinkhowIcouldgetfrominputto ambitiousgoals,theydeveloppowerfultechniquesthatof- output,sotheprogramcouldbemuchmoreunified,much ten have very relevant spin-offs. Even so, after they’ve smaller, and would also work. I mean, troff was collaps- solved that problem, they’regoing to think of something ingallthetime. AlotoftheearliestusersofTEXhadbeen elsewhichwillbeanotherrefinementandsoon—they’ll frustratedbytroff breakingoverandoveragain,soithad neverhaveasituationwherethey’regoingtoautomatically gottenunwieldy.Butithadalsoprovedthattherewaslight createthemostbeautifuldocument. There’sgoingtobea attheendofthetunnel. timewhenyoucanlookattheoutputandseethatyoucan stillimproveit. Designersofthemostautomaticsystems IalsohadtoscrapTEX,youknow,andstartoveragain;af- would be well advised to at least still leave a chance for terfiveyears,Idecidedthatitwouldbebesttogobackand somebodyto movesomething up and down and fake out re-do the program. But it would have been very hard to theirautomaticalgorithm. dothatif my friendin the nextofficehaddoneit[laugh- ter]. So, I just have this philosophythat therewill beal- ThephilosophythatIhadwhenIdidTEXwasthatIwould wayssomepeoplewhoaremoreinterestedinqualitythan trytohaveasystemthatdid99%ofeverythingautomati- others,andIwantedtomakeTEXgoodforthem. Idon’t cally,andthenIwouldlookatwhatremainedandIwould seeanygoodwaytomakeitimpossibletomakeabaddoc- kludgetherest. Butkludgingitisonewaytosayit;another ument,unlessyouhaveonlyasystemwith asmallmenu wayofsayingitis“Tidyuptherest,” or“Dotthei’sand ofoptions;that’sgoodforalargeclassofusers,tomakea cross the t’s.” My feeling is that this non-automaticpart systemthat’ssosimplethatyoucan’tpossibledoanything givesmealittleextrapridethatIhaveputthespitandpol- uglyinit. ishonthefinalproduct,thatIknowIdidit. Ifitoccursa lot,thenit’sanuisanceandI’mwastingtime. ButifIcan Erik: Ithinkit’stimeforacoffeebreaknow—we’lltake reallylimitthisto 1%—if I’vespent30hourswritinga fiveortenminutes. paperandittakesmeonlyanother15minutestocleanup, thenI’mhappytodoanother15minutesattheend. It’sa smalllittleextrathatgivesmeachancetocelebratethefact Knuth: Johannes,youhadaquestionthatyouhadtoask, thatI’vefinishedthepaper. solet’sgetthatoverwith[laughter]. ThespacingthatTEXdoesworstrightnow,inmyexperi- Johannes Braams: It’s abouttypesetting. What is your ence,iswithrespecttosquarerootsignsbeingalittletoo opinion about the skyline model of typesetting? In TEX, tight,withtheoperandeithertooclosetotheradicalsign you talk aboutboxes: each letter is inside a box, and we ortooclosetothebarlineorboth;IfindthatI’mmostof- glueboxestogethertoaline,andthelineitselfisinsidea ten fiddlingwith that. I’veadoptedin the bookConcrete box,andeachlineisviewedasaboxandtheboxesarefit- MathandalsoinTheArtofComputerProgrammingnow, tedtogethertoformaparagraph.Theskylinemodeltriesto theconventionwhereinthemathformulaIputan@-sign goalittlebitfurtherthantherigidboxandline,andtriesto where I want one math unit of extra space. The @-sign takeintoaccountthatsomeofthedescendersintheupper isthendefinedtohaveamathcodeofhexadecimal8000, lineandthehighpartsin thelowerlinedon’toverlap,so whichmeansthatthiswillinvokeinmathmodeandthe@- thatyoucouldactuallyhavelinesmuchtightertogether— sign will beregardedas a macro thatadds onemath unit especiallyinmathtypesetting,thatcouldbeanadvantage. ofspace. SoI’lltype‘squarerootof’‘@-sign’‘logofn’ [laughter],becauseotherwisethespacebefore‘el’isalittle Knuth: Hmmm,Iguessyou’retalkingaboutgeneralprin- bittootight. Nowmaybeeventhisskylinemodelwouldn’t ciplesofcomputergraphicswhereyouhaverectanglesin- knowthat‘el’wastootight,maybeitwould. Butit’scases side a picture, instead of having the rectangles grouped likethat... onlyinsideofarectangle....Thiscertainlywouldbeama- jorchangeinallthedatastructuresofTEX. Youcouldgoto Themostcommoncase reallyis whereI havesomething aquadtreestructureorsomethinglikethat. Allthethings like ‘x squared over 3’, where you have a simple super- thatpeopleusetosolvehidden-lineproblemsanddoren- scriptandthenaslash,andthenthedenominator. There’s dering, to find out what’s in frontof something else, and almostalwaystoomuchspacebeforetheslash. Andthisis allthealgorithmstheyusetomakemovieslikeToyStory. true,Ifind,inallthebooksthatIusedtothinkweretype- Itwouldbemostvaluable,Iimagine,forcatchingunusual setperfectlybyhand[laughter],butnowI’msensitiveto casesinmathformulas. this. Now I go through, typically with emacs, and look foralloccurrencesofsomethingwithaone-characterexpo- Ihavetwofeelingsaboutthesethings. OneisthatIliketo nentfollowedbyaslash,andmostofthoselookbetterwith seepeopleextendingthethingsthatcomputerscandoau- tomatically.Peoplelearnalotwhentheytrytodothis. The BijlageN KnuthmeetsNTGmembers 45 anegativethinspacebeforetheslash. ItwouldbenicerifI forthings,andoneofthethingswetrytoavoidinmathe- didn’thavetodothat. Butstill,it’sasmallthingforme. maticsisdoublesubscripts. IreadoneFrenchPhDthesis wheretheauthorhadfivelevelsofsubscripts[laughter]— Wouldtheskylinemodelhelpmemuch? SometimesIrun hekeptpaintinghimselfintoatrap. Hestartedoutwitha into cases where I’ll add another word to the answer to setx throughx ,sothenwhenhetalkedofasubset,ithad anexerciseinordertoavoidaclashbetweenlines. Here, 1 n tobexsub-i throughxsub-i ,andthenhewantedtotalk the lines are actually notgetting spread apart too far, but 1 k ofasubsetofthis,sothenhehadatheoremthatsays,leta they’resoclosetogetherthatthe‘subscriptkless-than-or- sub-bsub-c... andsoon[laughter]. Itrytochoosenota- equalton’willclashwithaleftparenthesisinthenextline. tionsthatgivemetheeconomyofthoughtatahighlevel. AndIdon’twantthetypetobequitesoclosetogetherthere. Now,ifIhadbeensmarter,Iwouldhavedesignedmy(cid:20)- That’s why I probably didn’t believe in a great deal of signtohaveadiagonalstrokeunderthe<insteadofahor- mark-upintheTEXbook;Iwouldbegintypewritertypeand izontalbar, and I wouldn’thavehad thoseclashes— too endtypewritertypeforsectionsbysaing\beginttand lateforthatnow.[laughter] \endtt. IwouldalsodelimitthelinesandwhenI’mpre- sentingpartsoftheplainTEXmacros,\beginlinesand Kees? \endlines—thosemacrosareinthefile,sinceit’svery Kees: May I ask you a question about your attitude to importanttometoseehowthatworks. Butinothercases, mark-upingeneral?Andletmeillustrateitbyfirsttellinga I’ve left [things] as simple as possible, for me to see vi- story. WhenwestartedwithusingTEXetc.,wemeanactu- suallythebeginningandendofstuff. It’ssomethingalso allywestartwithLATEX—Imean,thatistheeffectinHol- likeproblemsolving—sometimes,ifI’vesolvedaprob- land. AndthenIlookedattheproductsofthemark-upand lemandI’mnotworriedaboutitanymore,Iforgettotell Ididnotlikeit. AndthenIwaswondering,whatisyour anybodyelsethe solution. I wasalwaysaverybadcom- attitudeto that? I’m sorry to say so, I paged throughthe mitteechairmanbecauseI’mnotverygoodatfinishingthat TEXbookfiletexbook.texandIlookedatallthethings last endingline, I guess. Still, withhtml, thedocument inthereandthenIthought,“Well,Ihavesomeideaofwhat wasshortandIdecidedthatmyhomepagesweregoingto yourideasareofmark-up.”Andwhenyouexplainedabout beusedbymanydifferentkindsofbrowsingsoftwaresoI METAFONT and all those thingsnot in there, which you hadbetterbeveryrigorous. haveimplicit—amIwrongifIsummarizethis,thatyou WhileI was developingTEX, I attendedoneof themeet- adheretosomethinglikeminimalmark-up? ingsofthecommitteethatdesignedSGMLandhadavery gooddiscussionwithCharlieGoldfarbandtheotherpeo- Knuth: Yes. Forexample,whenIamreadingEdsgerDi- pleonthecommittee—weonlyhadthatonemeetingnear jkstra’sbooks, everytime I getto a section whereit says Stanford. CertainlyIappreciatethefactthatthisstructure ‘Endof Comment’, it strikes me as redundant. And I al- makesitpossibletobuildotherkindsofprogramsaround waysthink,“Oh,yes,thisisEdsger’sstyle.” WhenIwrote what you have. The more structure you have in a docu- apaperforhis60thbirthday,Isaidattheend,“Acknowl- ment,theeasieritistomakeadatabasethatincludesthings edgment,IwanttothankEdsgerforsuch-and-such,”and aboutit,andknowswhat’sgoingon. Ineverobjectedtoit; ‘EndofAcknowledgment’[laughter].4 Butthat’stheonly Ijustalwaysfeltthatinordertomaximizemyefficiency,I timeinmylifeI’lleverdothat. MaybeI’manillogicalper- didn’twanttomessaroundwithfullmark-upunlessIhad son,butapparentlyhalfthepeopleusinghtmlnowtype to. only thep at the beginningof a paragraph,and theother half type only a /p at the end of a paragraph [laughter]. X: SGML allows minimizations; that’s why the end- Hardlyanybodyusesboth,accordingtowhatmyspiestell paragraph is not necessary. So that’s one of the reasons me. AndIdon’tknowwhattheheckthesesystemsactu- whyit’ssodifficultsometimes. Youhaveaformalization allydowiththeunbracketedmaterial. WhenIwritehtml, tominimize. I’mscrupulouswithmymark-up. Ifyoulookatmyhome pages — I’ll pay you $2.56 if you find any case where I Knuth: ButLATEXdoesn’tallowit. started somethinganddidn’tcloseit with therighttag. I Johannes: Butwedohavesomebooks,however,permit- tried to be very careful in that, and to indent everything verywell,andsoon. ButIfounditaterriblenuisance,be- tingomittedend-tagsinLATEX3,butthat’snotfarenough along. causeit’snotthewayIthink. I think a high-level language, to me, is something that Knuth: Well,talktohim[laughter].Idon’tneedaspecial shouldreflectitsstructureinsomevisualwaybutnotnec- editorforhtml—peoplearehypingfancythingswhere essarily explicitly; so that, whenI knowtheconventions, youcanclickonatoolandit’llputinthestartandendtag we can suppress some things. Parentheses are one such together—butwhenIwrotemyfiles,Ididmakeupasim- conventionandmathematicsgota lotbetter whenpeople pleemacsmacrothatwouldtakewhatevertagIjusttyped inventedothernotationslikeoperatorprecedencethatwe andcreatetheend-tag. Allithadtodowassearchbacktill canseestructurewithoutspellingitoutintoomuchdetail. itfoundaless-thansignandthencopythatstringtwiceand A mathematicianspends a lot of time choosing notations 4BeautyisOurBusiness(Springer,1990),242. BijlageN KnuthmeetsNTGmembers 46 putaslashinfrontofit,soIusedthatallthetime—itwas theolddays,IwouldhavechangedtheplainTEXmacros easy. sothatitwouldnotusetheheightoftheminussigninthe \rightarrowfilloperator[...]6 Anyways,I’venow Johannes: Quite different type of question now, from toldyouthereasonwhyit’stherefortheotherones. someonewho’dliketoaskhere: literally,hewrites,“Why is theheightof theminussign in the cm symbolfontthe Johannes: Anotherquestion,whichisaboutmultiplelan- sameastheheightofthecmrplussign?” guages. There’saproblemwhenyouhaveoneparagraph whereyouhavedifferentlanguages. Knuth: Ah. Alotofpeoplearewonderingaboutthatone. Where youhave‘a minusc’ or yousay ‘x sub minus’or Knuth: Yes,the\lccodechanges. Thisisthe... something,whyis itthatthe heightanddeptharegreater Johannes: And I’vebeen told thatinside oneparagraph than the actualshape of the minussign?. In fact, it’s not just the plus and minus, it’s also the C, −, (cid:6), (cid:7), (cid:8), (cid:9), youcanonlyuseonehyphenationtable,whichistheone ⊗,(cid:11),(cid:2)and(cid:4)—ifyoulookatthecodeforthese,there whichisactiveattheendoftheparagraph. So,switching hyphenationtablesinsideparagraphs. Suppose,forexam- isabeginarithcharmacrothatbeginsallofthearithmetic ple,youhaveaparagraphwithEnglishtext,withaGerman charactersinthefont,guaranteeingthattheywillhavethe quoteinsideit,theGermanquotebeingseverallineslong. samesize. Johannes: Butitdoesn’tsaywhy. Knuth: IknowthatTEXwillproperlykeeptrackofwhich hyphenation table to use. The glitch, the mistake, that Knuth: That’sright—itdoesn’tsaywhy.Andthereason I didn’t anticipate is if the two languages have different \lccode mappings — so that each has a differentidea isthatearlyon,Iwantedcertainthingstolineupthesame. ofwhichcharactersarelowercase. Whenyouhyphenate, Forexample,ifyouhad youneedtohyphenateanuppercasewordthesameasan p p x C y C x − y; lowercaseword,soTEXusesthe\lccodeofacharacter toconverteveryletterintothelowercasecodeofthatlet- ter. Ididn’tanticipatethatpeoplemight,fordifferentlan- Iwantedthesquarerootsignstobeplaceinthesameway. guages,havea differentmappingfromuppercasetolow- Otherwise,youwouldget ercase. Andsoit’sthatmappingthat,attheendofapara- p p x C y C x − y: graph, applies to all the languagesin the paragraph. But otherwise, TEX is careful to keep track of what language [slightlyexageratedheretoshowthepoint]Andsothere youhave. are many other cases where you have formulas where Andbytheway,there’safilecalledtex82.bug. Goto there’saplussigninonepartofaformulaandaminussign the CTAN archives, and find subdirectory /knuth, and intheotherpart,andforconsistencyofspacing,itoughtto underthat/errata,andthat’swherethisis. Attheendof looksymmetrical. Thereareothercases, I readilyadmit, tex82.bugthisparticularerrorabout\lccodeismen- whereyouhaveonlyaminussign—youneverhaveasim- tionedasbeingsomethingthat’sanoversightthat’stoolate ilarthingwithaplussign,andyouwonderwhythere’sex- tofix. traspaceleftthere. SoIsay\smash minus[laughter] inthosecases. MarcvanLeeuwen: Why is it too lateto fix? It would conflictwithotherthings? Johannes: The particular application, why this question wasasked—MichaelDownesfromtheAMS— Knuth: Yes. Sothatpeoplearealreadyusingthesethings inlotsofdocuments,andit’sveryhardtochange.Infact,I Knuth: Yes, Michael Downes, he has more experience don’tseeanywaytofixit[laughter].Iwouldsaythatwhen thananyofusinthisroom;he’sthechieftypesetterofmost youarefacedwithasituationwhereyou’redoingmultiple ofthemathematicsintheworld. languageswithmultiple\lccodes,thisisagoodreason towriteyourownversionofTEX. Johannes: He hasaproblemproperlyattachingasuper- scriptontopofthe\rightarrowfill... Andries Lenstra: Could I ask a question? Happily enough, I’m not the first person to mention LATEX, so I Knuth: The \rightarrowfill? OK ... The may mentionit now. There’sa situation that often arises \rightarrowfillis this thing that makes a rightar- when people try to write a PhD thesis where they want row of any desired length, and then he wants to put a tochangeLATEXcodebecausetheythinkthey knowbet- superscript on this. What’s the macro for building that ter about things of beauty or typography, and unhappily up? Ihaven’tusedthatpageinalong... [laughter]5 The enoughtheyarenotexpertsonLATEX, so theydon’tsuc- \rightarrowfill is made up of minus signs and so ceedortheysucceedbadly. Ingeneral,peoplewhoknow probablyif I hadknownMichael... knownaboutthatin 5Knuthwastryingtoremember\buildrel;seeTheTEXbook,p.437. 6Infact,the\leftarrowfilland\rightarrowfillmacrosnowomittheheightanddepthoftheminus,inplain.tex version3.14159(March1995). BijlageN KnuthmeetsNTGmembers 47 abouttypographycan’twritebeautifulLATEXcodeorother andifnot,wouldyouthinkofgivingitabetterprogram- formsofcode,andviceversa—peoplewhoknowhowto minglanguage? write these forms of code, are no experts on typography. Knuth: InsomesenseIputinmanyoftheprogramming Whatdoyouthinkoftheendeavorsinthepasttobringthe featureskickingandscreaming,andI’lltrytoexplainthe two worlds together, for instance, as Victor Eijkhout has triedtodowithhislollipopformat,amachinetocreate background.IknowhowLesliewentaboutwritingLATEX — first he wouldwrite the algorithmsoutin a high-level otherformats. Iwouldhavethoughtthatitwouldhavehad programminglanguage,withwhile’sandif-then’sandso abigsuccessbuttheoppositeseemstobethetruth. What on, andthen he wouldpretty muchmechanicallyconvert doyouthinkofit? thistoTEXmacros. IfIhadsuspectedthatsuchastylewas Knuth: I’m notfamiliar with thedetailsoflollipop. goingtobethemostcommonuseofTEX,Iprobablywould IsupposethatwasbasedonafamousquotationfromAlan haveworried a lot in thosedays. Now, computersare so Perlis,whosaidthat,“Ifsomebodytellsyouthathewantsa fastthatIdon’tworrysomuchabouttherunningtime,be- programminglanguagewhichwillonlydotherightthing, causeitstillseemstogozip! givehimalollipop.” Inthe70s,Ihadanegativereactiontosystemsthattriedto Andries: Yes. beallthingstoallpeople. EverysystemIlookedathadits ownuniversalTuringmachinebuiltintoitsomehow,and Knuth: I’m sure that the lollipop effort was instructive everybody’s was a little different from everybody else’s. andworthwhile,butIdon’tknowthedetailssoIcan’tan- SoIthought,“Well,I’mnotgoingtodesignaprogramming swer in great detail on this. Probably the type designers language; I wanted to have just a typesetting language.” didn’tfindthelanguageeasytolearn. Idothinkthatwe’re Littlebylittle,Ineededmorefeaturesandsotheprogram- having much more communication now, as every month mingconstructsgrew. GuySteelebeganlobbyingformore goesby,betweenthepeoplethatknowabouttypeandthe capabilitiesearly on, and I putmanysuch thingsinto the peoplethatknowaboutmacros. It’sjustamatteroftime secondversionofTEX,TEX82,becauseofhisurging.That as we wait for these waves to continuemoving — we’re madeitpossibletocalculateprimenumbersaswellasdo nowherenearaconvergentstage, whereTEXhasreached complicatedthingswithpagelayoutandfigureplacements. its natural boundaryand the type designershave reached ButthereasonIdidn’tintroduceprogrammingfeaturesat their naturalboundary. They’restill movingtowardeach firstwasbecause,asa programmer,Iwastired ofhaving other. Idon’tthinkit’slikeahyperbolicgeometry,where to learn ten different almost-the-same programming lan- theyneverwillgettogether. guagesforeverysystemIlookedat;Iwasgoingtotryto avoidthat. Later, I realised that it was sort of inevitable, ThemaindifficultyofcourseisthatTEXisfree, andsoa butI triedtokeepitascloseto theparadigmofTEXasa lot of people will say, “Well, how could it be any good, character-by-charactermacrolanguageasIcould.AsIsaid ifyou’renotchargingmoneyforit?” Alotofthepeople before,Iwasexpectingthatthereallyspecialapplications in the typedesign communitywouldonly workin things wouldbedonebychangingthingsinthemachinelanguage wherethere’smoneybehindit;moneyprovestothemthat code. But peopledidn’tdo that, they wanted to putlow- it’s worth talkingto people. So it justtakes alittle while levelthingsinatahigherlevel. till they see some goodexamples, which will makethem moreopenforthesediscussions. Andthat’shappeningall Piet: Whatdoyouthink,forexample,ofsomethinglike thetimeindifferentcountries. buildinginaprogramminglanguagewhichis,fromasoft- In the Czech RepublicI was quite delighted to learn that wareengineeringpointofview,easiertouse? thenewencyclopaediainCzech,whichisthefirstonefor Knuth: Itwouldbeniceiftherewereawell-understood manyyears,isbeingdonewithTEX. Andnotonlythat,it’s standardforaninterpretiveprogramminglanguageinside beingdonewithaveryhighbudget. Theymadethisdeci- ofanarbitraryapplication. Takeregularexpressions—I sionbecausetheytriedalltheothersystemsandweredis- defineUNIX as“30definitionsofregularexpressionsliv- gustedwiththem. TheyhadgoodresultswithTEX. Many ing under oneroof.”[laughter]Every part of UNIX has a othercommercialpublishersareusingittoobecausethey slightly different regular expression. Now, if there were talktotheirfriendsatthebigpublishinghouses. Thiswill, auniversalsimpleinterpretivelanguagethatwascommon Ithink,besolvedwithtime. Andproductslikelollipop toothersystems,naturallyIwouldhavelatchedontothat areveryworthwhileinthemeanwhiletofacilitatethis. It rightaway. takestimetobringdifferentcommunitiestogether. Ithink thefinancialfactorisdefinitiveforalotofpeople. Piet: The Free Software Foundation is trying to do that andSunistryingtodoitandMicrosoftistryingto... PietvanOostrum: Idon’tknowifyouhaveeverlooked into theLATEX codeinside, butif youlookintothat, you Knuth: TheFreeSoftwareFoundationistryingactually gettheimpressionthatTEXisnotthemostappropriatepro- toincludealsothesolutionsofSunandMicrosoft. Inother gramminglanguagetodesignsuchalargesystem. Didyou words,tomakealloftheconventionsworksimultaneously everthinkofTEXbeingusedtoprogramsuchlargesystems asmuchaspossible. Andthatconflictswithmyownstyle, whereI’vetriedtohaveunityratherthandiversity...Idid-
Description: