MetPurity: A Learning-Based Tool of Pure Method Identification for Automatic Test Generation

In object-oriented programming, a method is pure if calling the method does not change object states that exist in the pre-states of the method call. Pure methods are widely-used in automatic techniques, including test generation, compiler optimization, and program repair. Due to the source code dep...

Full description

Saved in:
Bibliographic Details
Published in:2020 35th IEEE/ACM International Conference on Automated Software Engineering (ASE) pp. 1326 - 1330
Main Authors: Yu, Runze, Zhang, Youzhe, Xuan, Jifeng
Format: Conference Proceeding
Language:English
Published: ACM 01.09.2020
Subjects:
debugging
Error analysis
machine learning
Method purity
Random forests
regression testing
Software engineering
static analysis
test generation
Test pattern generators
Training
Training data
ISSN:2643-1572
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In object-oriented programming, a method is pure if calling the method does not change object states that exist in the pre-states of the method call. Pure methods are widely-used in automatic techniques, including test generation, compiler optimization, and program repair. Due to the source code dependency, it is infeasible to completely and accurately identify all pure methods. Instead, existing techniques such as ReImInfer are designed to identify a subset of accurate results of pure method and mark the other methods as unknown ones. In this paper, we designed and implemented MetPurity, a learning-based tool of pure method identification. Given all methods in a project, MetPurity labels a training set via automatic program analysis and builds a binary classifier (implemented with the random forest classifier) based on the training set. This classifier is used to predict the purity of all the other methods (i.e., unknown ones) in the same project. Preliminary evaluation on four open-source Java projects shows that Metpurity can provide a list of identified pure methods with a low error rate. Applying Met-purity to EvoSuite can increase the number of generated assertions for regression testing in test generation by EvoSuite.
ISSN:2643-1572
DOI:10.1145/3324884.3415292