EMOS Framework Article.docx
《EMOS Framework Article.docx》由会员分享，可在线阅读，更多相关《EMOS Framework Article.docx（18页珍藏版）》请在冰豆网上搜索。
EMOSFrameworkArticle
EMOSFramework:
FunctionalTestingBasedonGUI-Modelling
DeanRajovic
EMOSComputerConsultingGmbH
Oskar-Messter-Str.25
85737IsmaningbyMunich
Germany
ABSTRACT
InthispaperwediscusstheroleofGUI-modellingintheareaoffunctionaltestingandpresentaframeworkforthedevelopmentandexecutionofautomatedfunctionaltestsbasedonsuchmodelling.Themainissuesaddressedbytheframeworkare:
rapidtestdevelopment,intensiveinvolvementofnon-technicalpersonnelintestautomationandmaintainabilityofthetestenvironmentthatscaleswellforlargesystems.
ThemainfeaturesofourapproachtoGUI-modellingarereducingGUIoperationstotheirSET（enter）andCHK（verify）alternativesandcapturingthenavigationasanintrinsicpartofthetestdescription.Throughasimplelinkageconcept,anextremelyhighlevelofreusabilityoftestdataandtestcodeisachieved.
ThetechnicalimplementationisrealisedwiththepopulartesttoolWinRunner®fromMercuryInteractiveandisavailableundertheLesserGeneralPublicLicence（LGPL）.Theconceptisapplicabletopracticallyanyapplicationdesignedforintensivedatamaintenancevia"standard"GUIobjects.Ithasbeensuccessfullyappliedtoseveralhighlycomplextestautomationendeavoursinbrancheslikeinsurance,bankingandtelecommunications.
KEYWORDS:
testing,testautomation,functionaltesting,testframework,WinRunner
1
INTRODUCTION
Therearemanychallengesintestingcomplexsystems.Automatingtestsinsuchenvironmentsreducessomeofthembutintroducesadditionalones.Testswhicharesuitableformanualtestingarenotnecessarilysuitablefortestautomation（providedtheyhavebeenspecified/writtenatall）.Thesheeramountofpossibletestcasescanscareeventhemostexperiencedtestmanager/specialist.Especiallywhensuchtestsaretobecarriedoutwithhopelesslyunderstaffedtestteamsandwithinridiculoustimeframes...[5]Organisingnumeroustestartefactsintoamaintainablecomplexthatsurviveslongenoughtoproveitsusefulnessisaverychallengingtask[3,4,6,7,8].
Themostcommonapproachtofunctionaltestingisbyexercisingtheapplicationthroughitsuserinterface.AutomationoffunctionaltestssimplycannotignoretheGUI-orientedtesttoolsfromvendorssuchasMercuryInteractive,Rational,Segue,Compuware,etc.Althoughthesetoolsarecapableofimplementingverycomplexapproaches,advancedusersseemtobeleftalonetodiscovertheadvancedtechniquesontheirown.
ItdoesnottakemuchbrowsingthroughtheliteratureorintheInternettodiscoverthatmanypeoplefoundtheirownsolutions[1,2,3,8].Thesesolutionsalwaysseemtofollowthesamepath:
fromcapture/replay（record/playback）viascriptinganddata-driventestingtotestframeworks.Duetothelackoffirmtheoreticfundamentsthatcouldguaranteetheiruniversalapplicability（whichisprobablythereasonnoneoftheleadingtoolvendorsoffersuchframeworks）,theseframeworkspairtheirowntestingmethodologywiththetechnicalsupportforit.Nevertheless,theyalltrytosolvethesameproblem:
makethetestautomationmoreeffectiveandmoreefficient.
Whenitcomestotestingviatheapplication'suserinterface,theseframeworksappeartousethesameapproach:
eachmodelsthetestdesigninitsownwayandprovidesaninterfacetoone（ormore）standardGUI-centredtesttoolthatexecutesthetests.ThemostinfluentialtestframeworksofarisprobablytheCMG'sTestFrame®[1].Thisisthepowerfulframeworkbasedonakeyword-drivenseparation（intheirterminology:
"actionwords"）betweenthetestdata（"testclusters"）andthetestscripts.Themostimportantcommonpropertyofalmostalloftheapproachesatthismaturitylevelistheseparationofthetestdatafromthetestscripts.Themoreadvancedtheseparation,themoreadvancedtheconcept（e.g.simpledata-drivenversuskeyword-drivenapproach）.
2THE“BUSINESSCASESYNDROME”
Advancedconceptsattempttoexpressthetestcasesintermsofbusinessprocesses:
thehigh-levelabstractionsthathidetheiractualimplementationasmuchaspossible.Thepromisedbenefitsofsuch"businessmodelling"includetheindependenceoftheparticularuserinterface,increasedcomprehensiontonon-programmersandimprovedrequirements/designcoverage.
Howevertheseparationofscriptsanddatahasitscosts.Thecloserwegettothe"businesslanguage",thebiggerthegaptotheactualtestimplementation.Underthepremisethatweattempttoignoretheparticularuserinterface,itbecomesincreasinglymoredifficulttofindreusablecodepatternsthatrelatetosuch"businesslanguage".Suchpatternshoweverarefundamentaltobridgingthegapbetweenbusinessmodelsandtheirimplementation.
Businessprocessesarerarelysimple.Theproblemisoften"solved"bychunkingbigtransactionsintosmallerpartsand/orbyreducingtheinformationthathasactuallybeentested.Weoftenseefunctionslikeenter_xxx（...）,delete_xxx（...）andmodify_xxx（...）accompaniedwiththeclaimtheytestthegivenbusinessprocesses.Thequestionsiswhethertheyreallydo.
Thetestcodemustbeflexible.Amongotherthingsitshouldbeabletoenterandcheckvalidsituations,provokeerrors,takealternative,unusualorillegalpathsthroughtheapplication,handledynamic（context-sensitive）partsoftheapplicationandcheckthereactionoftheapplicationatmanydifferentplaces.Thepurposeofthiscodeistesting.Itshouldserveasa"playground"forimplementingtestideas.Itshouldbeabletomirrorthebehaviouroftherealhuman（anddestructive）testerasmuchaspossible.Thiscannotbeachievedwithacodeinwhichmanydecisionsarehardcoded.Whenanalternative"route"isneededorwhenthingsgetchanged（astheyinevitablydowithtesting）onemustmakeadaptationssomewheretogetthecodeworking.Questionsthatariseare:
∙Wheretoadapt?
∙Whocanadaptit?
∙Howmuchdoesitcost（time/money）toadapt?
Atestautomationexpert（testprogrammer,toolspecialist）ispermanentlyforcedtomakedecisionsaboutwhatinformationtomakepublic（i.e.makeita"businesscase"）andwhatinformationto"bury"intothetestscripts.Thetradeoffisthattheinformationwhichismadepubliccan（usually）bechangedbynon-programmers（domainexperts,endusers,testers）whereasthe"buried"informationrequiresprogrammingexpertisetochange.Simplechangesintheapplication'suserinterfacecaneasilyaffectplenty（hundreds）oftestcasesiftheinformationisexposedoronlyafewlinesofcodeiftheinformationishidden.Obviously,acarefulbalancemustbefound.However,itisagamble!
Moreoftenthanwewouldlike,changesaffectbothexposedorhiddeninformationregardlessofour"clever"balancing.Thosewhoknowthetricksoftestautomationwillappeartobe"luckier"thantheothers.
Themainbenefitoftestframeworksisinprovidingguidelinesforthosewhousethem（testdesigners,testprogrammers,endusers,systemdesigners,etc.）.Agoodtestframeworkprovidesgoodanswerstomanydifficultquestions.Theareathatstillrequiresplentyofresearchisinfindinganeffectivewayofexpressing（orevengenerating）testcasesthataresufficientlylegibleforhumansaswellasmachines.
3GUIHASITALL
Inourapproach,weconsciouslydigressfromthecommonlyacceptedpracticeofexpressingtestcasesinawaywhichis"independent"oftheparticularuserinterface.Tothecontrary,weconcentrateontheapplication'sinterfaceforthefollowingreasons:
∙itistheaspectthatnoneofthefunctionaltestautomationapproachescanignore,
∙itistheonlycommonplatformfordiscussionssuitabletoallparties,
∙itistheprimaryareaofinteresttomostexpertusersandtestmanagers,and
∙itisindeedsuitableforexpressing（functional）testcases.
Ourexperiencehasshownthatbymodellingtheuserinterfaceweareabletoexpressthe"businesscases"whilepreservingmostoftheotherbenefitsof"businessmodelling"（separationoftestsandscripts,increasedcomprehensiontonon-programmers,maintainability,scalability）andprovidefortheextremelyrapidandrobusttestautomation.Theapproachisparticularlywell-suitedforapplicationswithcomplexuserinterfacefocusedondatamaintenance.Thiscoversaverybroadrangeofbusinessapplicationsinpracticallyallbranches.
Thebasicobjectivesofourapproachare:
∙itmustbesetupquickly,
∙itmustrapidlyevolve（newtestcasesmustbeadoptedquickly）,
∙itmusteventuallyscaletoaccommodatehugesystems,
∙itmustbeableto"live"long（maintainableovermanyproductcycles）,and
∙itmustbeuseful（testwhatwasintendedtobetested）.
OurGUI-modellingideaisbasedonaprettyhereticalhypothesis.Ifthereisaneditfieldontheapplication'sGUI,allthatafunctionaltestwilldowithitistoenter,retrieve/checkorignoreitscontent.（Onemayarguethatthefieldattributes（label,position,enabled/disabled,etc.）alsoneedtobechecked.Thismaybetrue.However,comparedtotheoperationsonthefield'scontent,theoperationsonfield'sattributesarerare.Forthisreason,weconcentrateonthecontent.Nevertheless,wedoknowhowtohandleboth.）
ThesameholdstrueformostoftheotherGUIobjecttypes.Forexample,apushbuttonwillbepressedorignored,acheckboxwillbesetorcheckedwhetheritisset,elementsoflistboxeswillbeselectedorcheckedfortheirpresence,etc.Inotherwords,foreveryGUIobjecttypethereareonlyafewoperationsthatneedtobeappliedinordertoperformmostofthefunctionaltesting.
ThefirstmostimportantconceptofEMOSframeworkisthattheGUIofanapplicationisexercisedwithoperationsineitherSE
升级会员 