Covariant conversions (coco): A design pattern for type-safe modular software evolution in object-oriented systems

Publikation: Bidrag til bog/antologi/rapportKonferencebidrag i proceedingsForskningfagfællebedømt

Dokumenter

Software evolution is an essential challenge for all software engineers, typically addressed solely using code versioning systems and language-specific code analysis tools. Most versioning systems view the evolution of a system as a directed acyclic graph of steps, with independent branches that could be merged. What these systems fail to provide is the ability to ensure stable APIs or that each subsequent evolution represents a cohesive extension yielding a valid system. Modular software evolution ensures that APIs remain stable, which is achieved by ensuring that only additional methods, fields, and data types are added, while treating existing modules through blackbox interfaces. Even with these restrictions, it must be possible to add new variations, fields, and methods without extensive duplication of prior module code. In contrast to most literature, our focus is on ensuring modular software evolution using mainstream object-oriented programming languages, instead of resorting to novel language extensions. We present a novel CoCo design pattern that supports type-safe covariantly overridden convert methods to transform earlier data type instances into their newest evolutionary representation to access operations that had been added later. CoCo supports both binary methods and producer methods. We validate and contrast our approach using a well-known compiler construction case study that other researchers have also investigated for modular evolution. Our resulting implementation relies on less boilerplate code, is completely type-safe, and allows clients to use normal object-oriented calling conventions. We also compare CoCo with existing approaches to the Expression Problem. We conclude by discussing how CoCo could change the direction of currently proposed Java language extensions to support closed-world assumptions about data types, as borrowed from functional programming.

OriginalsprogEngelsk
Titel35th European Conference on Object-Oriented Programming, ECOOP 2021
RedaktørerAnders Moller, Manu Sridharan
ForlagSchloss Dagstuhl- Leibniz-Zentrum fur Informatik GmbH, Dagstuhl Publishing
Publikationsdato1 jul. 2021
Sider1-25
Artikelnummer4
ISBN (Elektronisk)9783959771900
DOI
StatusUdgivet - 1 jul. 2021
Begivenhed35th European Conference on Object-Oriented Programming, ECOOP 2021 - Virtual, Aarhus, Danmark
Varighed: 11 jul. 202117 jul. 2021

Konference

Konference35th European Conference on Object-Oriented Programming, ECOOP 2021
LandDanmark
ByVirtual, Aarhus
Periode11/07/202117/07/2021
SponsorAITO
NavnLeibniz International Proceedings in Informatics, LIPIcs
Vol/bind194
ISSN1868-8969

Antal downloads er baseret på statistik fra Google Scholar og www.ku.dk


Ingen data tilgængelig

ID: 296248521