Type relationship graphs for exploring APIs

• Autor(a) principal: Ferreira, Henrique Manuel da Costa Gomes
• Data de Publicação: 2017
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10071/14851
• topic: Engenharia da programação; Modelos de linguagem; Modelos gráficos; Análise estrutural; Software; Aplicação informática; API Usability; Code completion; IDE; Structural analysis; Usabilidade API; Sugestões de código

In the present day, the use of APIs, varying in complexity, size and design are pervasive, in the development of software. They are used to communicate, to log information, to draw charts and a myriad of other purposes. With this increasing importance in API usage, developers often struggle when using an API for the first time and are faced with various API discoverability problems. We propose an approach that aims to help mitigate the discoverability problems by providing code suggestions based on data extracted from structurally analyzing APIs for type relationships. The relationships are stored in a graph, which is used to navigate between the API types. The first steps when using an API are often one of the difficulties that developers face. For this reason, our approach provides API starting points to help the developer kick-start the use of the API. The next stage in API usage usually consists in using API types to access or create other API types. Our approach suggests possible type compositions based on the types available in the current development context. Depending on the API, the amount of extracted relationships can be overwhelming if directly suggested to the developer and for this reason a filtering mechanism and ranking heuristic were created, in order to provide more meaningful suggestions. Our approach was tested by analyzing 5 different APIs and simulating suggestions from the extracted data, comparing those suggestions to API usage examples. The results provide some evidence that our approach is a possible solution to the API discoverability problems and that the structural analysis to a given API can provide considerable amount of the information required to create and deliver suggestions to developers.