CodeBroker: An Active Reuse Repository SystemSoftware reuseWhy reuse?Reuse process (sLCMS)Research problemsUser’s knowledge about a reuse repositoryReinventing the wheelReinventing the wheelReinventing the wheelProposed solutionChallenges in active reuse repository systemsCodeBrokerInferring the taskSimilarity analysis in CodeBrokerThe rationaleBasic information retrieval (IR) techniquesLSA: Improved IRProbabilistic IR modelWeighting schema in CodeBrokerSignature matching determines the constraint compatibilitySignature matching for classesPresenter: tailoring the delivery to larger context and userDiscourse models: Improving task-relevanceUser models: User-specific deliveryIncremental discourse modeling and user modelingModels in CodeBrokerRetrieval-by-reformulationThe cycle ofdelivery-browsing-searchingSupporting comprehension and useEvaluating retrieval effectivenessEvaluation experimentsSystem assessmentRole of discourse modelsRole of user modelsProblems foundFuture researchGeneral lesson: Designing information repository systemsSummaryCodeBroker: An Active Reuse Repository SystemYunwen YeMar 15, 2004Software reuse Definition Creating new software systems with existing artifacts Reusable artifacts Code artifacts macros, functions, methods, classes, subsystems, systems Non-code artifacts analyses, designs, test plans and cases, domain models Knowledge program idioms, program plans, design patterns, software architecture styles, domain knowledgeReuse repository systems Supporting reuse activitiesWhy reuse? Increased productivity Reduced development time Reduced cognitive load Reduced testing time Increased quality Fewer bugs Enhanced evolvability and maintainabilityReuse process (sLCMS)Research problems No attempt to reuse (Location) Information islands Not aware of the existence of reusable components Perceived low reuse utility (benefits/cost) High cost of locating components Unable to locate the component (Location) Situation model vs. system model Unable to use the component (Comprehension)User’s knowledge about a reuse repositoryL3:BeliefL4:System ModelL2: VaguelyKnownL1: WellKnownReinventing the wheel Have you ever found that you have accidentally implemented a function that is in the library already? Countless times! (tomo) Yes this happens often while learning a new language. (prabhu) Yes, I have done this a number of times. (mandalia) Yup, I wrote a parser in Java that would have been much easier with a StringTokenizer. I'm sure I've done this other times, but that one really gets me (minick).  Yes. When I was trying to convert a string of numbers into integer, I wrote a function to do it. Later on I found out there is function atio in C library to the exactly the same thing (jing).  Probably many times, but how would I know? (Jon Marbach).Reinventing the wheel Have you ever found that you have accidentally implemented a function that is in the library already? Not yet (jackson) I cannot remember ever implementing a function that was already in the library. (deriggi) No, but I have never really checked this out (Serina Croll).Reinventing the wheel “Conversations with developers revealed several cases in which programmers, unaware of a virtual machine primitive for an operation, repeatedly reimplemented the same operation--in one case, ten times.” [Devanbu, 1991] Reusable objects demand proper advertisement [Walton, 1992] “We have discovered that ‘marketing’ the components in the CSL is just as important as providing the correct technologies for users in Schlumbeger Oilfield Services products. [Rosenbaum 1995] It happens that we develop functions when they exist and we do not realise it. [Coulange 1997] “I could be creating a method that does exactly the same thing somebody else’s does ... even though we have access to each other’s code. We might call them different names and we might have a bit different way of doing it, but we’re still doing the same thing.” [Fichman, Kemerer, 1997]Proposed solution Active component repository systems Overcoming the limits of browsing and searching Supporting information delivery Benefits Reusing unknown components Reduced locating cost Seamless integration with programming environmentChallenges in active reuse repository systemsL3:BeliefL4:System ModelL2: VaguelyKnownTask-relevant informationL1: WellKnownCodeBrokerEditing spaceExampleDelivery bufferIllustratorListenerListenerPresenterFetcherRepositoryinferred queriesretrieved componentsInferring the task Plan recognition Actions Æ Inferred goal Æ Suggested actions or information Similarity analysis Current situationCurrent situationneedssimilarInformation XSituation ASituation Aprobably needsSimilarity analysis in CodeBrokerCreate a random number between two limitsint <- int x intGenerate a random number using the default generatorint <- int x intSignature MatchingInformation RetrievalCurrent situationSituation AFetcherListenerThe rationale Three aspects of a program Concept The functionality of the program Semantic information Revealed in comments, identifiers, … Constraint Execution environment Syntactic information Revealed in signatures, protocols, … Code The implementation The assumption Similar concept + compatible signature Æ reusable codeBasic information retrieval (IR) techniques Information retrieval: Finding similar documents based on the commonality of terms Documents and queries are represented by term vectorsDj= (f1, j, f2, j, ..., fN, j)  Similarity is the distance between two vectorsSimilarity (Q, D) = Q[i]× D[i]i =1n∑Q[i]2i=1n∑× D[i]2i =1n∑Contents Vector SimilarityQ human factors in information retrieval system(1 1 0 1 0 1 1)D1 factor factor factor human human retrieval system(3 0 0 2 0 1 1) 7/750.5=0.80D2 information operation retrieval retrieval (0 1 0 0 1 2 0) 0.55D3 factor help help retrieval (1 0 2 0 0 1 0) 0.37Term space: (factor information help human operation retrieval system)LSA: Improved IR Latent semantic analysis Addressing the vocabulary mismatch problem (people use different names to refer to the same concept) Creating a semantic space with a large amount of documentsReducing the singular vectorsProbabilistic IR model Adding weights to each term Dj= (t1, j , t2, j, ..., tN, j) ti,j= TRWi* fi,j