ACM Turing Lecture 1972.docx

ACM Turing Lecture 1972.docx

《ACM Turing Lecture 1972.docx》由会员分享，可在线阅读，更多相关《ACM Turing Lecture 1972.docx（11页珍藏版）》请在冰豆网上搜索。

ACMTuringLecture1972

ACMTuringLecture1972

EWD340

TheHumbleProgrammer

by

EdsgerW.Dijkstra

AsaresultofalongsequenceofcoincidencesIenteredtheprogrammingprofessionofficiallyonthefirstspringmorningof1952andasfarasIhavebeenabletotrace,IwasthefirstDutchmantodosoinmycountry.Inretrospectthemostamazingthingwastheslownesswithwhich,atleastinmypartoftheworld,theprogrammingprofessionemerged,aslownesswhichisnowhardtobelieve.ButIamgratefulfortwovividrecollectionsfromthatperiodthatestablishthatslownessbeyondanydoubt.

Afterhavingprogrammedforsomethreeyears,IhadadiscussionwithA.vanWijngaarden,whowasthenmybossattheMathematicalCentreinAmsterdam,adiscussionforwhichIshallremaingratefultohimaslongasIlive.ThepointwasthatIwassupposedtostudytheoreticalphysicsattheUniversityofLeidensimultaneously,andasIfoundthetwoactivitiesharderandhardertocombine,Ihadtomakeupmymind,eithertostopprogrammingandbecomeareal,respectabletheoreticalphysicist,ortocarrymystudyofphysicstoaformalcompletiononly,withaminimumofeffort,andtobecome.....,yeswhat?

Aprogrammer?

Butwasthatarespectableprofession?

Forafterall,whatwasprogramming?

Wherewasthesoundbodyofknowledgethatcouldsupportitasanintellectuallyrespectablediscipline?

IrememberquitevividlyhowIenviedmyhardwarecolleagues,who,whenaskedabouttheirprofessionalcompetence,couldatleastpointoutthattheykneweverythingaboutvacuumtubes,amplifiersandtherest,whereasIfeltthat,whenfacedwiththatquestion,Iwouldstandempty-handed.FullofmisgivingsIknockedonvanWijngaarden'sofficedoor,askinghimwhetherIcould"speaktohimforamoment";whenIlefthisofficeanumberofhourslater,Iwasanotherperson.Forafterhavinglistenedtomyproblemspatiently,heagreedthatuptillthatmomenttherewasnotmuchofaprogrammingdiscipline,butthenhewentontoexplainquietlythatautomaticcomputerswereheretostay,thatwewerejustatthebeginningandcouldnotIbeoneofthepersonscalledtomakeprogrammingarespectabledisciplineintheyearstocome?

ThiswasaturningpointinmylifeandIcompletedmystudyofphysicsformallyasquicklyasIcould.Onemoraloftheabovestoryis,ofcourse,thatwemustbeverycarefulwhenwegiveadvicetoyoungerpeople;sometimestheyfollowit!

Anothertwoyearslater,in1957,ImarriedandDutchmarriageritesrequireyoutostateyourprofessionandIstatedthatIwasaprogrammer.ButthemunicipalauthoritiesofthetownofAmsterdamdidnotacceptitonthegroundsthattherewasnosuchprofession.And,believeitornot,butundertheheading"profession"mymarriageactshowstheridiculousentry"theoreticalphysicist"!

SomuchfortheslownesswithwhichIsawtheprogrammingprofessionemergeinmyowncountry.SincethenIhaveseenmoreoftheworld,anditismygeneralimpressionthatinothercountries,apartfromapossibleshiftofdates,thegrowthpatternhasbeenverymuchthesame.

Letmetrytocapturethesituationinthoseolddaysinalittlebitmoredetail,inthehopeofgettingabetterunderstandingofthesituationtoday.Whilewepursueouranalysis,weshallseehowmanycommonmisunderstandingsaboutthetruenatureoftheprogrammingtaskcanbetracedbacktothatnowdistantpast.

Thefirstautomaticelectroniccomputerswereallunique,single-copymachinesandtheywerealltobefoundinanenvironmentwiththeexcitingflavourofanexperimentallaboratory.Oncethevisionoftheautomaticcomputerwasthere,itsrealisationwasatremendouschallengetotheelectronictechnologythenavailable,andonethingiscertain:

wecannotdenythecourageofthegroupsthatdecidedtotryandbuildsuchafantasticpieceofequipment.Forfantasticpiecesofequipmenttheywere:

inretrospectonecanonlywonderthatthosefirstmachinesworkedatall,atleastsometimes.Theoverwhelmingproblemwastogetandkeepthemachineinworkingorder.Thepreoccupationwiththephysicalaspectsofautomaticcomputingisstillreflectedinthenamesoftheolderscientificsocietiesinthefield,suchastheAssociationforComputingMachineryortheBritishComputerSociety,namesinwhichexplicitreferenceismadetothephysicalequipment.

Whataboutthepoorprogrammer?

Well,totellthehonesttruth:

hewashardlynoticed.Foronething,thefirstmachinesweresobulkythatyoucouldhardlymovethemandbesidesthat,theyrequiredsuchextensivemaintenancethatitwasquitenaturalthattheplacewherepeopletriedtousethemachinewasthesamelaboratorywherethemachinehadbeendeveloped.Secondly,hissomewhatinvisibleworkwaswithoutanyglamour:

youcouldshowthemachinetovisitorsandthatwasseveralordersofmagnitudemorespectacularthansomesheetsofcoding.Butmostimportantofall,theprogrammerhimselfhadaverymodestviewofhisownwork:

hisworkderivedallitssignificancefromtheexistenceofthatwonderfulmachine.Becausethatwasauniquemachine,heknewonlytoowellthathisprogramshadonlylocalsignificanceandalso,becauseitwaspatentlyobviousthatthismachinewouldhavealimitedlifetime,heknewthatverylittleofhisworkwouldhavealastingvalue.Finally,thereisyetanothercircumstancethathadaprofoundinfluenceontheprogrammer'sattitudetohiswork:

ontheonehand,besidesbeingunreliable,hismachinewasusuallytooslowanditsmemorywasusuallytoosmall,i.e.hewasfacedwithapinchingshoe,whileontheotherhanditsusuallysomewhatqueerordercodewouldcaterforthemostunexpectedconstructions.Andinthosedaysmanyacleverprogrammerderivedanimmenseintellectualsatisfactionfromthecunningtricksbymeansofwhichhecontrivedtosqueezetheimpossibleintotheconstraintsofhisequipment.

Twoopinionsaboutprogrammingdatefromthosedays.Imentionthemnow,Ishallreturntothemlater.Theoneopinionwasthatareallycompetentprogrammershouldbepuzzle-mindedandveryfondofclevertricks;theotheropinonwasthatprogrammingwasnothingmorethanoptimizingtheefficiencyofthecomputationalprocess,inonedirectionortheother.

Thelatteropinionwastheresultofthefrequentcircumstancethat,indeed,theavailableequipmentwasapainfullypinchingshoe,andinthosedaysoneoftenencounteredthenaiveexpectationthat,oncemorepowerfulmachineswereavailable,programmingwouldnolongerbeaproblem,forthenthestruggletopushthemachinetoitslimitswouldnolongerbenecessaryandthatwasallwhatprogrammingwasabout,wasn'tit?

Butinthenextdecadessomethingcompletelydifferenthappened:

morepowerfulmachinesbecameavailable,notjustanorderofmagnitudemorepowerful,evenseveralordersofmagnitudemorepowerful.Butinsteadoffindingourselvesinthestateofeternalblissofallprogammingproblemssolved,wefoundourselvesuptoournecksinthesoftwarecrisis!

Howcome?

Thereisaminorcause:

inoneortworespectsmodernmachineryisbasicallymoredifficulttohandlethantheoldmachinery.Firstly,wehavegottheI/Ointerrupts,occurringatunpredictableandirreproduciblemoments;comparedwiththeoldsequentialmachinethatpretendedtobeafullydeterministicautomaton,thishasbeenadramaticchangeandmanyasystemsprogrammer'sgreyhairbearswitnesstothefactthatweshouldnottalklightlyaboutthelogicalproblemscreatedbythatfeature.Secondly,wehavegotmachinesequippedwithmulti-levelstores,presentingusproblemsofmanagementstrategythat,inspiteoftheextensiveliteratureonthesubject,stillremainratherelusive.Somuchfortheaddedcomplicationduetostructuralchangesoftheactualmachines.

ButIcalledthisaminorcause;themajorcauseis...thatthemachineshavebecomeseveralordersofmagnitudemorepowerful!

Toputitquitebluntly:

aslongastherewerenomachines,programmingwasnoproblematall;whenwehadafewweakcomputers,programmingbecameamildproblem,andnowwehavegiganticcomputers,programminghadbecomeanequallygiganticproblem.Inthissensetheelectronicindustryhasnotsolvedasingleproblem,ithasonlycreatedthem,ithascreatedtheproblemofusingitsproducts.Toputitinanotherway:

asthepowerofavailablemachinesgrewbyafactorofmorethanathousand,society'sambitiontoapplythesemachinesgrewinproportion,anditwasthepoorprogrammerwhofoundhisjobinthisexplodedfieldoftensionbetweenendsandmeans.Theincreasedpowerofthehardware,togetherwiththeperhapsevenmoredramaticincreaseinitsreliability,madesolutionsfeasiblethattheprogrammerhadnotdaredtodreamaboutafewyearsbefore.Andnow,afewyearslater,hehadtodreamaboutthemand,evenworse,hehadtotransformsuchdreamsintoreality!

Isitawonderthatwefoundourselvesinasoftwarecrisis?

No,certainlynot,andasyoumayguess,itwasevenpredictedwellinadvance;butthetroublewithminorprophets,ofcourse,isthatitisonlyfiveyearslaterthatyoureallyknowthattheyhadbeenright.

Then,inthemid-sixties,somethingterriblehappened:

thecomputersoftheso-calledthirdgenerationmadetheirappearance.Theofficialliteraturetellsusthattheirprice/performanceratiohasbeenoneofthemajordesignobjectives.Butifyoutakeas"performance"thedutycycleofthemachine'svariouscomponents,littlewillpreventyoufromendingupwithadesigninwhichthemajorpartofyourperformancegoalisreachedbyinternalhousekeepingactivitiesofdoubtfulnecessity.Andifyourdefinitionofpriceisthepricetobepaidforthehardware,littlewillpreventyo