26/03/14

Dont Repeat Yourself

DontRepeatYourself
Context: Duplication(inadvertentorpurposefulduplication)canleadtomaintenancenightmares,poorfactoring,andlogicalcontradictions. Duplication,andthestrongpossibilityofeventualcontradiction,canariseanywhere:inarchitecture,requirements,code,ordocumentation.The effectscanrangefrommisimplementedcodeanddeveloperconfusiontocompletesystemfailure. OnecouldarguethatmostofthedifficultyinY2Kremediationisduetothelackofasingledateabstractionwithinanygivensystemthe knowledgeofdatesanddatehandlingiswidelyspread. TheProblem: Butwhatexactlycountsasduplication?CloneAndModifyProgrammingisgenerallycitedasthechiefculprit(seeOnceAndOnlyOnce,etc.),but thereismoretoitthanthat.Whetherincode,architecture,requirementsdocuments,oruserdocumentation,duplicationofknowledgenotjusttext istherealculprit. Therefore: TheDRY(Don'tRepeatYourself)Principlestates: Everypieceofknowledgemusthaveasingle,unambiguous,authoritativerepresentationwithinasystem. It'sokaytohavemechanical,textualduplication(theequivalentofcachingvalues:arepeatable,automaticderivationofonesourcefilefromsome metaleveldescription),aslongastheauthoritativesourceiswellknown. Forexample,inamixedlanguageCORBAenvironmentyoumaychoosetotreattheIDLdefinitionsasauthoritative.Thesedefinitionswillbe usedtoautomaticallygeneratesourcecodefileswhichduplicatetheknowledgeintheIDL.Butthat'sokay:theIDLformoftheknowledgemeets therequirementsofourdefinition. Wheredifferentlevelsofabstractionareinvolved,aconsistentconflictresolutionschememustbeused.Thiscouldbeassimpleasidentifyingone levelasauthoritativeinallcases,oralwaysdeferringtothehighlevel,orwhatever,aslongasitisconsistentlyapplied. Forexample,inC++theinterfaceandimplementationforaclassaretypicallyspecifiedinseparatefiles,duplicatingknowledge.Youmayconsider theheaderfiletobeauthoritativeforthecontractoftheclassasviewedbyitsclients,andthesourcecodetobeauthoritativeregardingissuesof implementationwhicharehiddenbytheimplementation. Notes:ThisprincipleissimilartoOnceAndOnlyOnce,butwithadifferentobjective.WithOnceAndOnlyOnce,youareencouragedtorefactorto eliminateduplicatedcodeandfunctionality.WithDRY,youtrytoidentifythesingle,definitivesourceofeverypieceofknowledgeusedinyour system,andthenusethatsourcetogenerateapplicableinstancesofthatknowledge(code,documentation,tests,etc). Theprinciplealsoassumesthatyourprojectshaveahighdegreeofautomation,allowingthegenerationofthederivativeknowledgeartifacts wheneverrequired. AndrewHunt ThefollowingconversationsgetmuchsimplerwhenwerecognizethemostimportantthingtoDRYupis... declarationsofbehavior KeepingthestructureofaprogramDRYisharder,andlowervalue.It'sthebusinessrules,theifstatements,themathformulas,andthemetadata thatshouldappearinonlyoneplace.TheWETstuffHTMLpages,redundanttestdata,commasand{}delimitersarealleasytoignore,so DRYingthemislessimportant. PhlIp TheoppositeisWET: We Edit Terribly,Tumultuously,Tempestuously,Tenaciously,Toomuch,Timidly,Tortuously,Terrifiedly... IthinkWETalsostandsfor"WeEnjoyTyping" DuncanBayne

c2.com/cgi/wiki?DontRepeatYourself

1/7

26/03/14

Dont Repeat Yourself

But: Itmaytaketimeandefforttoselectand/orcreateadefinitivesourceseeYouArentGonnaNeedIt.DickBotting Hmm."Everypieceofknowledgemusthaveasingle,unambiguous,authoritativerepresentationwithinasystem."Isthisnotareasonable definitionforaSingleton? No.DRYreferstoyoursourcecode,notyourrunningprogram.ThirdNormalFormistheanalogousprinciplefordata. Ugh.Ithinkyou'rebothbarkingupthewrongtreesthere.Itdoesn'trefertosourcecode,nortotherunningprogram.Itreferstoasystem.Needn't havecomputersinvolvedatall. InTheArtOfUnixProgramming,EricRaymondcallsthistheSinglePointOfTruthrule(SPOT). Iagreewiththis,"inprinciple".However,takenliterally,thisprincipleisimpossibletoachieve.Forexample,afunctionsignatureisduplicatedat thefunctioncallsiteandthefunctiondefinitionsite.Evenifyoudon'tdeclaretypes,thenumberofargumentsisafactthatisduplicated.Ipreferto thinkofgettingridofduplication(i.e.OnceAndOnlyOnce)insteadofneverhavingitinthefirstplace,becauseyoukeepdiscoveringthat somethingthatyoudidn'tthinkwasaduplicate,reallyis.Also,sometimesitismoretroublethanitisworthtoeliminatesomeduplication. Moreover,itisoftenbettertowaittilyouhaveseveralexamplesoftheduplicationsoyoucanbetterseehowtoeliminateit.RalphJohnson Ralph,that'sareallygoodwayoflookingatthings.Mybiggestprogrammingheadacheshavealwayscomefromabstractlystrugglingwith"how canIwriteagoodgeneralsolutiontothisproblem",eventhoughIonlyknowofoneplacewhereit'sdefinitelygoingtobeused."I'dbetterthink ofageneralsolutionnow,"Ithink,"orI'llhavetocopyandpasteandchangecode!"Butit'sabsurdtotrytocomeupwithageneralsolution withoutknowingmoreaboutthedifferentvarietiesoftheproblemthatexist(orwillexist)inthesystem. It'sabattleoftworeallystrongurgesOnceAndOnlyOncevsavoidingPrematureGeneralization.DoIduplicatefornowandtrytolivewiththe duplicationforawhile,orviolateYagNiandcomeupwithsomehalfcockedgeneralizedsolution?It'satoughone,becausealmostall programmershateduplicationit'sasortofprimordialprogrammingurge. However,eventhoughCopyAndPasteProgrammingcanbeexpensivetocleanup,soisabotched"generalsolution"andcopyandpasteisfar cheaperupfront.SoIalsoaminfavouroftemporaryduplication,toberefactoredwhenyouhaveaclearerviewofthesituation. MatthewBennett Alas,justexactlyhowIthoughtaboutit.Nowthetemporaryduplicationisahellishpermanentduplication.Ihavenotimetoremoveit,andthe moreIhavetoupdatetheeviltwinsofcode,thelesstimetoeverdosomethingaboutitiget.L.A.Nilsson (Formoreintheveinof"letaduplicationliveforawhile,inordertoseehowtobesteliminateit",seeCodeHarvesting.) IthinkthereisalsoacollisionwithDoTheSimplestThingThatCouldPossiblyWorkwhereIbelievethatlattershouldwinmostofthetime(aside fromthesituationswhererefactoringistoocomplicated,databaseschemas?) TheDRYprincipleisreallyaphilosophicalissue(althoughitdoesrearitsheadretroactivelyduringcoding).Theideaistotrytoplanaheadto preventduplication,ratherthattowastetimeremovingstuffyou'vealreadydone.Functionsignaturesareaformofduplication,albeitafairlyminor one,asthecompilershouldbeabletospottimeswhereonesidegetsoutofstep.It'sthemoreinsidiousduplicationwehopetostopwithDRY. Dave
' ' W h yi st h i sd u p l i c a t i o n ?N e i t h e rt h ep r o t o y p ea l o n en o rt h ef u n c t i o nd e f i n i t i o nc a r r i e s t h ef u l lm e s s a g e .W i t h o u tt h ep r o t o y p ee r r o r sw i l lc r e e pi n ,a n dw i t ho u tt h ef u n c t i o nd e f i n i t i o ne r r o r s c r e e pi n .T h i si sn o td u p l i c a t i o ni ti sv i r t u o u sr e d u n c a n c yw h i c hi n c e a s e st h et o t a le r r o rt r a p i n gp o w e r s y n e r g i s t i c l y .D i dIm i s st h ep o i n ts o m e w h e r e? ' '

Ilikethisprinciple.Perhapsrelatedtotheideaofbeingabletotraceanypieceofcodingbacktothereasonsforitsexistenceandform.Formea keypointistheselectionofoneitemasadefinitivesourceforcertainpurposes. Forexample,asintheabovedescription,I'dnormallytakeafunctionprototype/header/declarationasguidingthefunctioncalls.ThisiswhyIlike headerfilestocontaincommentsdescribingthecontract(makesurethefirstargumentpointsatrealstorageandthiscallwilldo.....foryou).The functionbodiesaredefinitivefortheimplementationdecisionsbutmightbederivedfrommoredefinitivegenericalgorithmsperhaps? DickBotting IfcodedoesnotconformtotheDRYprinciplecanwesaythatitiswet? Icertainlylikethisprinciple.Theapplicationsinembeddedsystemsshouldbeobvioustothecasualobserver.Canthepumpdispensemedication ornot?Onlyoneentityknowstheabsoluteanswertothispuzzler.Everycomponentthatneedstoknowwhatthemachine'scurrentstatusisgoes toawellknownmethod(aresource)tofindout.Whatunitsarewedealingwith,mlormg/cc?Eachcomponentmayneedalocalcopyofthis settingtofigureouthowtoconvert,buttheyallgotoacentralrepositorytoacquirethatknowledge.Mostexcellent,wouldn'tyousay? Let'stryasimpleXPcodingexample:

c2.com/cgi/wiki?DontRepeatYourself

2/7

26/03/14

Dont Repeat Yourself

areallthesesqrtexamplesreallymakingthepoint?Shouldwedeletethemormovethem?AnonymousCoward Agreedthesquarerootexamplekindawentoffonatangent,thediscussionitselfwasn'ttoobadbutitdoesn'tbelonghereAndrewRicketts{be boldaboutyouropinions} P1:OK,weneedasquarerootfunction.Letscodethefirsttestcase

t e s t 1 :a s s e r t _ e q u a l ( s q r t ( 4 ) ,2 ) ;

P2:andthecode:
s u bs q r t ( x ){r e t u r n2 ;}

P1:Hmm.OK,perhapsweneedanothertesttogetabetterimplementation
t e s t 2 :a s s e r t _ e q u a l ( s q r t ( 9 ) ,3 ) ;

P2:OK
s u bs q r t( x ){i f( x = = 4 ){r e t u r n2}e l s e i f( x = = 9 ){r e t u r n3}}

P1:Youcan'tdothat,you'rejustrepeatingtheinformationinthetest:itviolatestheDRYprinciple P2:Thatdoesn'tmatter.CodemustonlyobeyOAOO.DRYisjustphilosophicalclaptrap.Thetestsdon'tprovideenoughinformationto generalize,sothesimplestthingistoduplicatetheinformation. P1:No,itdoesmatter.Atestanditscodemustusedifferentalgorithmswhenexpressingthesameinformation.Iftheyarethesamethenthetest addsnovalue. Whoiscorrect?Whatimplementationwouldenablethetests+codetoconformtoDRY? ''Here'smyresolutionforthisparticularexample: Startwithagoodstatementabouttheprocesstodesignandtest. Documentedrequirement: 1. 2. 3. 4. 5. 6. 7. producethesquarerootof(x). Wheresqrt(x)*sqrt(x)=x. Errororimaginarynumbersmustbereturnedforx<0. x=0shouldreturn0. Insertedrequirement:xisintheRealNumberDomain. Insertedperformancerequirement:FunctionOverflowmustbeconsideredandreportedasanerror. Insertedperformancerequirement:AsFastAsPossible/DoNotBottleNeck.

Thenwriteyourtest:
f o r ( y = 1t o1 0 0 ){ x=r a n d o m _ i n _ r a n g e ( 1 ,s o m e _ m a x )-w h yn o tr e p l a c e1b ys o m e _ m i n-p o s s i b l yl a r g ea n dv e ? a s s e r t _ e q u a l ( x , s q r t (( x * x )) )

youmightalsowanttoconsiderassert_equal(x,sqrt(x)*sqrt(x))whichisadifferentproposition.DavidMartland
} a s s e r t _ e r r o r _ o r _ i m a g i n a r y ( s q r t ( 1 ) ) a s s e r t _ z e r o ( s q r t ( 0 ) )

noticethattherequirementdoesnotspecifyanupperdomain.Somodifiedthetestforrandomvalues>0.ThiswayMr.GuruCoderdoesn't getlazyandimplementanycomplexifstatements.Ifstatementnothing.Aguardclauseandatablelookup.MrGuruCoder.Ihappento knowfromexperiencethatsuchatabletendstogetexpensivewith32bitandlargerintegers(severaloptimizationsandIcouldn'tgeta PrimeNumbertablebelowaquartergigbeforegivingupbuthey,whataweekendofcoding). finallycodeandtestthesqrtfunction. WyattMatthews [Notedinpassing:Thistestdoesn'tseemtodomuchtowardverifyingthatthecodeworkswhentheargumenttosqrtisnotaninteger.Indeed,ifx isafloatingpointnumberthatisnotaninteger,thesquarerootofxcanneverberepresentedexactlyasafloatingpointnumber(Counterexample: sqrt(0.25)=0.5isperfectlyrepresentableinbinaryfloatingpointRV)(becausethesquarerootofanonintegerisalwaysirrationaldefinitely wrong:sqrt(0.01)==0.1,where0.01isnonintegerand0.1isrationalMrMathematician).Sotestingasquarerootprogramoutsidetheinteger domainisamuchharderproblem.AndrewKoenig]

c2.com/cgi/wiki?DontRepeatYourself

3/7

26/03/14

Dont Repeat Yourself

[Secondnoteinpassing:ImisspokewhenIsaidthatthesquarerootofanonintegerisalwaysirrational.WhatImeanttosayisthatthesquareroot ofanintegeriseitheranintegerorirrational.Now,anyfloatingpointvaluexcanbemultipliedbyanevenpowerof2toyieldaninteger.Callthe smallestsuchnumberN.Thensqrt(x*N)==sqrt(x)*sqrt(N),andbecauseNisanevenpowerof2,sqrt(N)isapowerof2.Becausex*Nisan integer,sqrt(x*N)iseitheranintegerorirrational.Ifsqrt(x*N)isaninteger,thensqrt(x)isanexactfloatingpointnumber,anditcanberepresented withhalfthesignificantbitsofx.Ifsqrt(x*N)isirrational,thensqrt(x)mustalsobeirrational,soitcannothaveanexactfloatingpoint representation. SowhatI'mtryingtosayisthatifyoutakethesquarerootofafloatingpointnumber,theneithertheresultisexact,andhaslotsofloworder zeroes,oritisanecessarilyinexactapproximationtoanirrational.AndrewKoenig] [Theassertionthatsqrt(x)canberepresentedwithhalfthesignificantbitsofxisnotalwayspreciselycorrect.] Thankstothecommentinsertedandtheblockbelow,Iwillnote,thatIdidn'twritethecodeforthesqrtfunctionasimpletablemaystillsufficeto bypassmydomainlesstestasitmaybeproventhatsqrt(x)=sqrt(y)*sqrt(z).Anyoneuponaproofthatthismight(ormightnot)work?Also, insteadofdoingastupidintegralsearch(asshownbelowitisonlystupidbecauseitdoesnotcheckfory*y>xandbreak)abinarysearchofthe RealNumbersDomainwillatleastaffordsomeadditionalspeed. WyattMatthews Ifxisaninteger...
i n ts q r t ( i n tx ){ i n ty=1 ; w h i l e( y * y! =x )+ + y ; r e t u r ny ; }

I'dliketoseeaproofthatthiswillhaltwhenxisanoddnumberotherthan1. Here'syourproof:Ifx=9,then:y=1,y*y=1y=2,y*y=4y=3,y*y=9.returnsy=3 OKbutitdoesn'thaltifxisoddandalsonotasquare.DavidMartland Moreappropriately,youshouldhaveaskedforaproofthatthefunctionwillpasstherandomtestifx=random_in_range(1,some_max)can specifynonintegralvaluesorvalues<=0withrespecttotherequirement.Asitstands,thisfunctionisnotcodedtosurvivealargeportionofthe IntegralDomain,muchlesstheRealNumbersDomain.Thisfunctionshouldberenamedtosomethingsuchas"IsXaPerfectSquare?"andmodified alittle. Attheleast,thestatedrequirementwants: error_codesqrt(intx,doubleresult){

e r r o r _ c o d er e s ; d o u b l et e s t ; i f ( x < 0 )r e s=e r r o r _ o r _ i m a g i n a r y ; e l s e { r e s=n o _ e r r o r ; i f ( x = 0 ) r e s u l t=d o u b l e ( 0 ) ; e l s e { / / S o m ec o d et os e a r c ht h er e a ln u m b e rs p a c e . / / Ab i n a r ys e a r c ht oac e r t a i nd e p t hf o re x a m p l e . } } r e t u r nr e s ;

} WyattMatthews IfyouprogramitinCstyle,atleastmakeitworkinC(passvariablesbyvalue,etc) TheSingleAuthorityaffordedbytheDRYprincipleandtheduplicationofthesourceasabasisfortheprogramenvironmentaddsinertiatothe system(seeRedundancyIsInertia)forthesakeofitsstability.Here'salittlelistthatIbelieveshowshowRedundancyIsInertia,theDRYprinciple, OnceAndOnlyOnce,andtestingshouldworktogether: thesingleauthorityistheinterface'ssignatureanddocumentedrequirements/goals.(RequirementsAndGoalsareanotherissue) theauthorityneedstobereadablebyanypotentialdirectuser.(ie:atextualorgraphicalrepresentationforthesoftwarewriterandabinary representationforthesoftware) theinterface'sbody(theprocess)mayduplicatethesignatureifandonlyiftheimplementationenvironmentrequiresit.

c2.com/cgi/wiki?DontRepeatYourself

4/7

26/03/14

Dont Repeat Yourself

TheprocessisafactItmustabidebyOnceAndOnlyOnceandtesttrueagainsttheauthority. R&Gmarkersmaybeplacedinthetextualcopyoftheprocessmuchaseditnotesaresometimesplacedthere. theuserduplicatesonlythepartofthesignaturerequiredtoestablishcommunications.(thecompilationprocessisallowedtoduplicatethe interface'sprocess.) theuseroftheinterfaceshouldonlyneedtesttheinterfaceagainsttheR&Gonceperlifecycle.(seeCategoryTesting) WyattMatthews Thereisapatterntotheexceptionstotheprinciple.Itisoktohavemorethanonerepresentationofapieceofknowledgeprovidedaneffective mechanismforensuringconsistencybetweenthemisengaged. DefinitionsanddeclarationsofCfunctions:theseareusuallyinsyncbecausethecompilerflagsinconsistencies,forcingtheprogrammerto takeaction. Unittests:inconsistencymeansthetestswillfail,againforcingsomeonetotakeaction. Autogeneratedstuff:periodicregenerationensuresconsistency. Oftentheeffectivemechanismisautomatedbutitdoesn'tneedtobe.Anexampleisenduserdocumentation:anautomatedmechanismfor verifyingsoftware/documentationconsistencyishardtocomeby(AiComplete?),soyouneedahumantoreviewthis.Amechanismtoensure consistencycouldinthiscasebeaninstructiononwhattodobeforeeachsoftwarerelease. AndersMunch DRYdoesn'tjustcombatinconsistency,though.Yes,apassedtestshowsthatcodeandtestareconsistent,buttheymaystillbothbewrong.Ifthe testandthecodehavethesameunderlyingstructure/knowledge,it'smorelikelythatadoublefailwillmakeapass.However,inmanyfunctions, it'seasytoeliminatethedoubleerrorsbecausetheirinverseisradicallydifferent.Forexample,considererrorcorrectingcodes:encodingisjusta hashwhiledecodingisacomplexoperation.Implementingtheencoderasatestreliablyfindserrorsinthedecoder. DRYsaysthatoneshouldstrivetouseadifferentimplementationintestandcodetoavoidthesameerrorinbothcausingafalsepositive.This actuallybringsupanotherpointwhatdoyoudowhentheonlywayofimplementingsomethingisidenticaltothewayyoutestit?Imustadmit I'mneverquitehappywhenIhavetodothat.Luckily,itusuallyonlyoccurswhentheoperationissosimplethatitbarelyneedstestingatall. JamesKeogh Seehttp://blogs.computerworld.com/but_it_seemed_like_such_a_good_idea_at_the_timeforanamusingdemonstrationoftheevilsofduplicate code. AristotlePagaltzis Postingalinkisitselfanamusingdemonstrationoftheevilsofduplicatecode(thislinkisavalidexample).MJones I'mstruckbythefactthatthisistheprinciplethatpushespeopletonormalizerelationaldatabases.It'snotwithoutitsdownsideconsideringthe reasonswhypeoplechoosestarschemas.ThereneedstobesomediscussionofwhenitisOKtoabandonthisrule.meta@pobox.com Actually,normalizeddatabasesrelyondataduplicationandoneusuallyhastowriteasetofconstraintstoensurethatkeyslinkingtablesare consistent.
H U H?T h ef a c tt h a tt h es a m e d a t ai si n t h eF Ka n dt h eP Ko ft h el i n k e dt a b l e si sn o td u p l i c t i o ni ti sa n e c e s s a r ym e t h o do fi m p l e m e n t i n gt h el i n k .

Meta'scommentaboutdatabasesremindsmeofthefactthattheschemaofthedatabaseandthecodethataccessthedatabasearetwo representationsofdatastructures.Ideally,boththecodeandthedatabaseschemawouldbedrivenoffthesamesourceofdatastructureinformation. ThisistheattractionofRuby'sActiveRecordandthereasonIhateHiberateandStruts.BothHiberateandStrutshaveduplicateinformationthat needstobeinsync.WithoutsomethinglikeXDoclettheywillgetoutofsyncandwillviolateDRY.SaS Merelyhopingtolearnwhat"DRY"is,Icame.Thethoughtinspiringphilosophicaldiscussionwastoomuch,merelytoread. Epidemiology(inshort,thestudyofdiseaseanddiseasedistributioninpopulations)promulgatestheideathatagivendiseasehassinglecausea rootcause.Insystems,RootCauseAnalysisinevitablytakesusbacktoadesignerrorandthisisoftentraceabletoanerrorintherequirements definition.(DatafromtheSoftwareEngineeringInstitute(SEI)atCarnegieMellonUniversity(CMU)showsanumberover60%,forthe percentageoferrorscausedwithinthestepspartofRequirementsDefinition.) Earliercomputerengineeringconcepts(say,early1980s)hadmaturedsufficientlythatstudents(andmanagers)mightbelievepeopleotherthan thosewhoareantisocialcanbeasgood(orbetter)thana"guru".Thecominggenerationofengineers/programmerswouldvaluewritingtheir ideaseveniftheydidn'tworkforIBM.Somemightevenplanahead.Thisalloweddesigntoavoidcircumstantialredundancy.TheDoDfunded workatCMUthatgaverisetotheSEIinthe1980srevealedthat"bestpractices"beginwith"planning"ofsomesort. Tosumupwhatcaneasilyoccupyvolumes:Weneedtolearnfromourmistakes.Thisrequiresconsciouseffortandtimecommitment,which makesitrare.Noacronymcanrepresenteveryproject,soacheatsheetmightsimplyreadas,"DRY".Thisisredundantwiththestartingpostbut Ididn'twriteitwhichmeansDRYhasnotbeenviolated.(Thoughthiscontrib/maybe/abitdry.)

c2.com/cgi/wiki?DontRepeatYourself

5/7

26/03/14

Dont Repeat Yourself

Kernel.package OnceAndOnlyOnceishardtoapplywhenconsumersappreciatePutThingsWhereYouLook.Andconsumerslookforeverythingeverywhere. Translation,"Whyisn'tthisemailalinkIcanclick.Youremailshouldbeonthecontactspage,theuserpage,thehelppage,andthehomepage." LeeLouviere

I tm u s tb ev i s a b l ee v e r y w h e r e .I tc a nl i v ei n o n l yo n ep l a c e .Av i e wi sn o tt h ed a t a .

CustomerrequeststrumpDRY.Ifsourcecodeontheinsideisdry,thenthoseemailclickerswillallusethesamecode,triviallypluggedinto severaldifferentviewlocationsPhlIp Hmm.Letmerephrase.PutThingsWhereYouLook(appliedtotheconceptofputtingthingseverywheretheusermightlook),mayleadto accidentallybreakingOnceAndOnlyOnce,andcanaccidentallybeviewedascontrasttoDontRepeatYourself.However,youshouldreally implementaprogramdataasuncoupledfromtheinterface.Don'tconfuseDRYtomeanthatinformationcanonlyhaveonelocationasopposedto onesource. Itaketheapproachtodesigndataandinterfaceasiftheywereseparateprograms,thenhavetheir"communication"derivedlast.ThatwayIensure thatdataisimplemented"native"tothemachinelanguageofsimplicity,andtheuserinterfaceisdesignedasmostconformingtohowtheuser woulduseitmeettheuserdemandofPutThingsWhereYouLookandthemyriadofdifferentarrangementsofthedatatoaccommodateeveryusers distinctwayofunderstandingdata.(EditMeneedabettertermhere,nativeisbad,communicationishorrible.)LeeLouviere Ifthisprincipleexistsatallitismorelikelytobecalled'eliminateneedlessredundancy'than'donotrepeatyourself.'A=Aisnotaredundant statement.Aisnotneedlessrepetition.A=BisnotthesameasA=AandAbyitselfdoesnotalwaysmeanA=Aifitistestandnotananalytic truth.Therearenumerousinstancesofclaimsofrepetitioninthediscussionabovewhereremovingoneoftheinstanceoftheallegedduplication willchangethemeaningandbehaviorofasystem.Ifyoucannotremoveoneoftheduplicateswithoutchangingmeaningitisnotduplication. Thenotionthatnormalizationrequiresduplicationbecausethekeysmustexistinbothtables,suggestthatremovingoneofthedubsispossiblebut itisnot. EvenIfIconcededthatduplicationisabettertermthanredundancythekeyisthatitisonlyneedlessduplicationthatmustbepurged.Theacidtest forneedless,forthosewhoseintuitionisnotasufficientguideisthis.Ifyoucan'tremoveitwithoutloosingfunctionalityitisnotneedless.
-A n n o n

OK,butalotoffunctionalityshouldstillbegeneralizedinmanycasessincethey'reverysimilar,andtherecouldbeabetter(moreelegant) implementationofit.Anonymous Sometimesmarketrulesdictatethechoiceofimplementationlanguage,whichcanmakeuseofDontRepeatYourselfextremelydifficult.Saythere's oneapplicationthatyouwanttodeployonmultipleplatforms,eachofwhichrequires(oratleastverystronglyprefers)codewritteninaparticular language.OneplatformrequiresObjectiveCee.Anotherplatformrequires100%pureJavaVirtualMachinebytecode.Athirdplatformrequiresthe useof100%verifiablytypesafeCIL(CommonLanguageRuntimebytecode).AfourthplatformrequiresJavaScript.ToachieveDRYunderthese constraints,onewouldhavetosomehowcreateaDomainSpecificLanguagethatcompilestoObjectiveC,compilestoJVMbytecode,compilesto safeCIL,andcompilestoJavaScript,andthenwritetheauthoritativeimplementationofthebusinesslogicinthatlanguage.Shouldeachcross platformapplicationprojectbeginbydesigningandimplementingaDSLasaworkaroundtoallowWriteOnceRunAnywheredespitethesemarket rules? (Forthosewhopreferconcreteexamples:iOSneedsObjectiveC,JavaMicroEditionphonesandJavaAppletbrowsersneedJavabytecode, WindowsPhone7andXboxLiveIndieGamesonXbox360needCIL,andFlash,HTML5browsers,andChromeWebStoreneedvariationsof EcmaScript.)DamianYerrick? DontRepeatYourself,butrepeatingothersisperfectlyacceptable:) Everypieceofknowledgemusthaveasingle,unambiguous,authoritativerepresentationwithinasystem. ThissoundslikesomethingoutofOrwell's1984.Anystatementthatbeginswith"Everypieceofknowledgemusthave.."soundsliketheofficial propagandaofatotalitarianstatenotthefreeexchangeofideas.Perhapstheprinciplereallybeingexpressedhereis"It'smywayorthehighway." {"...andshouldpayme$130k/yrtopoliceandenforcethispolicy.":} IactuallywholeheartedlyagreewiththisandamprettysureitiswhyItakesolongtogetanythingdone.Ispendmoretimetryingnottoduplicate myselfandfindingcleverwaystoabstractoutthevariousduplicatedflowprinciplesinmyapplicationthanactuallysimplycoding.Sometimes whenIignoreruleslikeyouhavestated,Igetaninsaneamountofworkdone...veryquickly...andthenIworryifitismessy. RichardEllicott See:UnnecessaryHolography,OnceAndOnlyOnce,DuplicationRefactoringThreshold

c2.com/cgi/wiki?DontRepeatYourself

6/7

26/03/14

Dont Repeat Yourself

CategoryModelingLawsAndPrinciples,CategoryExtremeProgramming,CategoryPlanning,CategorySimplification EditTextofthispage(lasteditedMarch21,2014)orFindPagewithtitleortextsearch

c2.com/cgi/wiki?DontRepeatYourself

7/7