- author = "Rodrigo {Morales Alvarado}",
- title = "Automated Improvement of Software Design by Search-Based Refactoring",
- school = "Departement de genie informatique et genie logiciel, Ecole Polytechnique de Montreal",
- year = "2017",
- address = "Canada",
- month = dec,
- keywords = "genetic algorithms, genetic programming, Genetic Improvement, SBSE, Software maintenance, design quality, anti-patterns, refactoring, search-based software engineering, energy efficiency, task context, testing effort, NSGA-II, SPEA2, MOCell",
-
URL = "
https://publications.polymtl.ca/2878/",
-
URL = "
https://publications.polymtl.ca/2878/1/2017_RodrigoMoralesAlvarado.pdf",
- size = "175 pages",
-
abstract = "Software maintenance cost is estimated to be more than
70percent of the total cost of system, because of many
factors, including new user's requirements, the
adoption of new technologies and the quality of
software systems. From these factors, quality is the
one that we can control and continually improved to
prevent degradation of performance and reduction of
effectiveness (a.k.a. design decay). Moreover, to stay
competitive, the software industry has shortened its
release cycles to deliver new products and features
faster, which results in more pressure on developer
teams and the acceleration of system's design
evolution. One way to prevent design decay is the
identification and correction of anti-patterns which
are indicators of poor design quality. To improve
design quality and remove anti-patterns, developers
perform small behaviour-preserving transformations
(a.k.a. refactoring). Manual refactoring is expensive,
as it requires to (1) identify the code entities that
need to be refactored; (2) generate refactoring
operations for classes identified in the previous step;
(3) find the correct order of application of the
refactorings generated, to maximize the quality effect
and to minimize conflicts. Hence, researchers and
practitioners have formulated refactoring as an
optimization problem and use search-based techniques to
propose (semi)automated approaches to solve it.
In this dissertation, we propose several approaches to tackle some of the major issues in existing refactoring tools, to assist developers in their maintenance and quality assurance activities. Our thesis is that it is possible to enhance automated refactoring by considering new dimensions: (1) developer's task context to prioritize the refactoring of relevant classes; (2) testing effort to improve testing cost after refactoring; (3) refactoring's conflict awareness to reduce refactoring effort; and (4) energy efficiency to improve energy consumption of mobile applications after refactoring.
We propose four approaches: (1) ReCon, which leverages developer's task context to prioritize the refactoring of classes that are relevant to the developer's activity. Using ReCon, developers can remove a median of 50percent of anti-patterns during regular coding tasks, without disrupting their workflow. (2) RePOR, for an efficient refactoring scheduling, which results in a reduction of refactoring effort and execution time by 80percent. (3) TARF controls for the testing effort while refactoring. Results show that TARF can reduce a median of 48percent of the testing effort of a system after refactoring. (4) EARMO, is an automated approach for the refactoring of mobile applications, which is able to remove 84percent of anti-patterns and extend the battery life of devices by up to 29 minutes (for a multimedia app running continuously a typical scenario).
We apply and validate our proposed approaches on several open-source systems to demonstrate their impact on design quality using well known quality models, and feedback from some authors of the systems studied.",
-
notes = "Is this GP?
Supervisors: Foutse Khomh and Giuliano Antoniol and Francisco Chicano",
