In this thesis, we present mixsets, an approach to combine annotative and compositional fragments for specifying code variants to form software product lines (SPLs). There are three key contributions of our research: introducing mixsets to represent software variability, extending mixsets to construct feature models, and improving software composition to achieve fine-grained variability. The concept of mixsets is introduced in Umple as a conditional unit and a first-class entity to allow smoothly transitioning software to compositional SPLs. A mixset is a named set of mixins; each mixin belonging to the mixset is called a fragment. A mixset fragment can be a top-level entity that contains nested entities or can be embedded as a conditional fragment in other entities such as methods. Mixset content normally includes code blocks or statements of any type, and may include require statements, which describe explicit dependencies among mixsets. Mixsets can be used to specify product lines using code composition, code annotation or both. A strength of mixsets lies on the straightforward mechanism to transform annotated segments into compositional segments when used in a combined approach. Therefore, preplanning effort and time to transform annotative SPLs to compositional SPLs can be reduced. Mixsets can provide a backbone structure to realize product line features in a feature-based SPL. Hence, a feature model can be formed using a subset of specific dependent mixsets. Feature interactions within an SPL can be identified and separated in specific modules by mixset nesting. Furthermore, product configuration, feature modeling analysis and generation of feature diagrams can be accomplished based on mixsets. The thesis also demonstrates a method to enable the granularity of compositional approaches to be expressed at the statement level. This is achieved by allowing aspects to inject code into labelled places, or points of variation, within method code. Such injected code has a full access to the context in which its placed, such as access to local variables. Mixsets are implemented in Umple, a model-driven development (MDD) technology that allows combining abstract models, such as associations and state machines, with pure code, and generating code in multiple programming languages. Mixsets can thus be used to describe variations in models, not just code. The thesis uses a design science approach as its research methodology. The mixset concept is evaluated through three case studies. The first applies mixsets to the Oracle Berkeley Database SPL, which is used in other literature to evaluate SPL concepts. This study shows that the mixset implementation improves on the state of the art in several respects, such as elimination of complex workarounds that are required by other approaches. In the second study, mixsets help to refactor Umple into feature-oriented SPL. The case study shows how annotative fragments can be easily transformed into compositional counterparts. In the third study, mixsets help to present two alternative solutions the to the Rover Control Challenge Problem of the MDETools 2018 workshop.

La réutilisation est reconnue comme une démarche intéressante pour améliorer la qualité des produits et la productivité des projets logiciels. L'ingénierie des logiciels à base de composants (CBSE en anglais) et l'ingénierie des lignes de produits logiciels (SPLE en anglais) sont considérées parmi les plus importants paradigmes de réutilisation systématique. L'ingénierie à base de composants permet de développer de nouveaux systèmes logiciels par composition de briques préconstruites appelées composants. L'ingénierie des lignes de produits logiciels permet de dériver (construire) de nouveaux produits par simple sélection de leurs caractéristiques (feature en anglais). Cette dérivation est rendue possible grâce à la représentation et à la gestion de la variabilité et de la similarité des produits d'une même famille. Cependant, une des difficultés vers une large adoption de l'ingénierie des logiciels à base de composants et des lignes de produits est le coût à investir pour construire, à partir de rien, les composants et les artefacts de lignes de produits. Dans ce contexte, les travaux de cette thèse proposent de réduire ce coût par une démarche basée sur la rétro-ingénierie.La première contribution de cette thèse consiste à proposer une approche permettant d'identifier, par l'analyse d'un ensemble de variantes d'un produit logiciel orienté objet, les éléments du code source pouvant servir à l'implémentation de composants. Au contraire des approches existantes d'identification de composants basées sur l'analyse d'un seul produit, l'originalité de notre approche consiste en l'analyse de plusieurs variantes de produits en même temps. Notre objectif est l'amélioration de la réutilisabilité des composants extraits. L'évaluation expérimentale menée dans le cadre de cette thèse a montré la pertinence de cette démarche.La deuxième contribution consiste en la proposition d'une approche pour l'extraction d'une architecture à base de composants d'un ensemble de variantes d'un produit logiciel orienté objet. Il s'agit d'identifier la variabilité des composants architecturaux et la configuration architecturale. L'identification de la configuration architecturale est principalement basée sur l'utilisation de l'analyse formelle de concepts pour trouver les dépendances entre les éléments architecturaux. L'expérimentation conduite pour l'évaluation de l'approche proposée confirme la pertinence des éléments identifiés.La troisième contribution de cette thèse permet de restructurer les APIs orientées objet en composants. Nous exploitons la spécificité des classes des APIs d'être conçues pour être utilisées par des applications clientes pour identifier ces composants. Le code source de ces APIs et celui de leurs applications clientes sont analysés afin d'identifier des groupes de classes qui peuvent servir à l'implémentation de composants à extraire. L'identification de ces groupes de classes est basée sur l'analyse des liens structurels entre ces classes et sur la probabilité que ces classes soient utilisées ensemble par les applications clientes. Nous montrons à travers les résultats expérimentaux que la restructuration des API orientées objet en composants facilite la réutilisation et la compréhension des éléments de ces APIs. / It is widely recognized that software quality and productivity can be significantly improved by applying a systematic reuse approach. In this context, Component-Based Software Engineering (CBSE) and Software Product Line Engineering (SPLE) are considered as two important systematic reuse paradigms. CBSE aims at composing software systems based on pre-built software components and SPLE aims at building new products by managing commonalty and variability of a family of similar software. However, building components and SPL artifacts from scratch is a costly task. In this context, our dissertation proposes three contributions to reduce this cost.Firstly, we propose an approach that aims at mining reusable components from a set of similar object-oriented software product variants. The idea is to analyze the commonality and the variability of product variants, in order to identify pieces of code that may form reusable components. Our motivation behind the analysis of several existing product variants is that components mined from these variants are more reusable for the development of new software products than those mined from single ones. The experimental evaluation shows that the reusability of the components mined using our approach is better than those mined from single software.Secondly, we propose an approach that automatically recovers the component-based architecture of a set of object-oriented software product variants. Our contribution is twofold: the identification of the architectural component variability and the identification of the configuration variability. The configuration variability is based on the identification of dependencies between the architectural elements using formal concept analysis. The experimental evaluation shows that our approach is able to identify the architectural variability.Finally, we propose an approach that aims at restructuring object-oriented APIs as component-based ones. This approach exploits specificity of API entities by statically analyzing the source code of both APIs and their software clients to identify groups of API classes that are able to form components. Our assumption is based on the probability of classes to be reused together by API clients on the one hand, and on the structural dependencies between classes on the other hand. The experimental evaluation shows that structuring object-oriented APIs as component-based ones improves the reusability and the understandability of these APIs.

[pt] Microsserviços é uma técnica industrial para promover melhor escalabilidade e manutenibilidade de pequenos e autônomos serviços. Estudos prévios sugerem que a arquitetura de microsserviços vem sendo amplamente usada para reduzir limitações encontradas em sistemas monolíticos legados tais como melhoria de inovação, uso de diferentes tecnologias, entre outras. O processo de migração para a arquitetura de microsserviços não é trivial. Este é particularmente o caso da tarefa de identificar candidatos a microsserviço e o código fonte associado com cada candidato que é dispendiosa e propensa a erro. Consequentemente, abordagens automatizadas têm sido propostas para reduzir o esforço relacionado a essa atividade. Essas abordagens comumente adotam um ou dois critérios para suportar a identificação de microsserviços com base no sistema monolítico legado. Contudo, existe uma falta de compreensão da utilidade desses critérios adotados na prática. Além disso, há limitado conhecimento em quais são os critérios que profissionais consideram relevantes. Levando em consideração esses limitantes existentes, nós conduzimos um survey e entrevista para melhor compreender a utilidade de critérios relatados em estudos empíricos (e.g, estudos de caso e relatos) do ponto de vista dos profissionais. Os resultados do survey e da entrevista mostram que as abordagens automatizadas e ferramentas existentes não são totalmente alinhadas com necessidades práticas. Para atender às necessidades deles, este trabalho define uma abordagem automatizada chamada toMicroservices. A abordagem baseia-se em uma combinação de análise estática e dinâmica do código legado. A abordagem visa indicar os candidatos a microsserviço e a fonte correspondente extraído do sistema legado. toMicroservices faz uso da engenharia de software baseada em busca para otimizar e balancear os cinco critérios comumente adotados por profisionais, nomeados de modularização de funcionalidade, redução de sobrecarga de rede, reúso, acoplamento e coesão. Além disso, um estudo de caso e grupo focal foram conduzidos a posteriori para avaliar e melhorar toMicroservices. / [en] Microservices is an industrial technique to promote better scalability and maintainability of small and autonomous services. Previous studies suggested that microservice architectures have been widely used to reduce limitations found in legacy monolithic systems such as the inclusion of innovation, use of a different stack of technologies, among others. The process of migrating to a microservices architecture is far from trivial. This is particularly the case for the task of identifying candidate microservices and the source code associated with each candidate, which is recognizably time-consuming and error-prone. Thus, automated approaches have been proposed to reduce the effort related to that task. These approaches commonly adopt one or two criteria to support the identification of microservices from a legacy monolithic system. However, there is a lack of understanding on the usefulness of these criteria in practical settings. Moreover, there is limited knowledge on what are the criteria that practitioners consider relevant. Taking into account these existing limitations, we conducted a survey and interviews to better understand the usefulness of criteria reported in empirical studies (e.g, case studies and reports) from the point of view of practitioners. The results of the survey and interviews revealed that existing automated approaches and tools are far from being aligned with practical needs. To fulfill these needs, this work defines a automated approach named toMicroservices. The approach relies on a combination of static and dynamic analysis of the legacy code. The approach aims at indicating the microservice candidates and the corresponding source extracted from the legacy system. toMicroservices makes use of search-based software engineering (SBSE) to optimize and balance the five criteria commonly adopted by practitioners, namely feature modularization, network overhead reduction, reuse, coupling and cohesion. Additionally, an industrial case study and a focus group were conducted a posteriori to support the evaluation and improvements of toMicroservices.

Une Ligne de Produits Logiciels (LPL) supporte la gestion d’une famille de logiciels. Cette approche se caractérise par une réutilisation systématique des artefacts communs qui réduit le coût et le temps de mise sur le marché et augmente la qualité des logiciels. Cependant, une LPL exige un investissement initial coûteux. Certaines organisations qui ne peuvent pas faire face à un tel investissement, utilisent le « Clone-and-own » C&O pour construire et faire évoluer des familles de logiciels. Cependant, l'efficacité de cette pratique se dégrade proportionnellement à la croissance de la famille de produits, qui devient difficile à maintenir. Dans cette thèse, nous proposons une approche hybride qui utilise à la fois une LPL et l'approche C&O pour faire évoluer une famille de produits logiciels. Un mécanisme automatique d’identification des correspondances entre les « features » caractérisant les produits et les artéfacts logiciels, permet la migration des variantes de produits développées en C&O dans une LPL. L’originalité de ce travail est alors d’aider à la dérivation de nouveaux produits en proposant différents scenarii d’opérations C&O à effectuer pour dériver un nouveau produit à partir des features requis. Le développeur peut alors réduire ces possibilités en exprimant ses préférences (e.g. produits, artefacts) et en utilisant les estimations de coûts sur les opérations que nous proposons. Les nouveaux produits ainsi construits sont alors facilement intégrés dans la LPL. Nous avons étayé cette thèse en développant le framework SUCCEED (SUpporting Clone-and-own with Cost-EstimatEd Derivation) et l’avons appliqué à une étude de cas sur des familles de portails web. / A Software Product Line (SPL) manages commonalities and variability of a related software products family. This approach is characterized by a systematic reuse that reduces development cost and time to market and increases software quality. However, building an SPL requires an initial expensive investment. Therefore, organizations that are not able to deal with such an up-front investment, tend to develop a family of software products using simple and intuitive practices. Clone-and-own (C&O) is an approach adopted widely by software developers to construct new product variants from existing ones. However, the efficiency of this practice degrades proportionally to the growth of the family of products in concern, that becomes difficult to manage. In this dissertation, we propose a hybrid approach that utilizes both SPL and C&O to develop and evolve a family of software products. An automatic mechanism of identification of the correspondences between the features of the products and the software artifacts, allows the migration of the product variants developed in C&O in an SPL The originality of this work is then to help the derivation of new products by proposing different scenarios of C&O operations to be performed to derive a new product from the required features. The developer can then reduce these possibilities by expressing her preferences (e.g. products, artifacts) and using the proposed cost estimations on the operations. We realized our approach by developing SUCCEED, a framework for SUpporting Clone-and-own with Cost-EstimatEd Derivation. We validate our works on a case study of families of web portals.