ApacheBugLearning

ApacheBugLearning is a Java project structured in Maven modules, capable of retrieving all commits of any Java project from the Apache Software Foundation, linked with related Jira tickets, and through machine learning predicting classes affected by BUGs.

git Module

The git module takes care, through the use of the Eclipse JGit project, of retrieving all the conmmits, related releases and files inherent to each commit of each release.

The source repository can be pre-existing locally, thus indentified by an absolute path or be cloaned locally automatically, indicating the choice within the configuration .properties file.The projects that can be cloned automatically are: TAJO, STORM, SYNCOPE, ZOOKEEPER, OPENJPA, AVRO and BOOKKEEPER.

During reading, several metrics are calculated for each file and entered into a .csv file. The metrics used:

Metric	Description
LOC	Number of lines of code
LOC ADDED	Sum of LOCs added
LOC MAX ADDED	Maximum LOCs added
LOC TOUCHED	Sum of LOCs added and LOCs removed
NUMBER OF REVISION	Number of revisions of the class
AVERAGE LOC ADDED	Average number of LOCs added
NUMBER OF AUTHORS	Number of reviewers of the class
CHURN	| LOC added - LOC removed |
MAX CHURN	Maximum CHURN
AVERAGE CHURN	Average CHURN
NPM	Number of public methods in the class
NPVM	Number of private methods in the class
NSM	Number of static methods in the class
NAM	Number of methods in the class
NLOCM	Number of commented lines

Project test files are discarded; you can un-discard them by setting the flag in the configuration file.

jira Module

The jira module deals with the management of the data recuoperated through the git module, in particular, it deals with associating each commit (object of type RevCommit) with the related Jira ticket, associating the commits and calculated metrics with the related files, and associating the files with the reference relaease.

Ticket information is retrieved in groups of 1000 tickets, via the Jira REST API, using the query:

"https://issues.apache.org/jira/rest/api/2/search?jql=project%20%3D%20" + projectName +
"%20AND%20issuetype%20%3D%20Bug%20AND%20(%22status%22%20%3D%22resolved%22%20OR%20%22status" +
"%22%20%3D%20%22closed%22)%20AND%20%20%22resolution%22%20%3D%20%22fixed%22%20" + "%20ORDER%20BY%20key%20ASC" +
"&fields=key,resolutiondate,versions,created,fixVersions&startAt=" + i + "&maxResults=" + j

All tickets that refer to BUGs, and that have been closed as a result of the Bug fix, are considered. The variables i and j are used to retrieve tickets in groups of 1000 items (group of maximum elements returned by a Jira REST API call).

A separate entity has been created for each of these items:

   commit → Bug Entity

   release → Release Entity

   file → RepoFile Entity

   repository → Repo Entity

Inconsistent tickets are discarded and if not present the injection version, necessary for class labeling, is calculated using proportion techniques (link paper). Under a threshold, set within the configuration file, proportion Cold Start is used otherwise porportion Increment.

dataset Module

The dataset module deals with the prediction of Bugginness of classes, using the training set constructed with the previous modules and as a Machine Laerning tool WEKA.

The linear combination of different techniques and classifiers was used for prediction. The Classifiers used were Random Forest, Naive Bayes, IBk in combination with Feature Selection Best First techniques (Backward Search and Forward Search), Sampling Oversampling, Undersampling and SMOTE techniques, and Cost Sensitive Threshold and Learning techniques.

For the evaluation of the classifiers, the Walk Forward technique was used; all the given data were divided into k Releases, ordered chronologically, and for each release a run was performed. At the k-th run, the k-th release was used as the testing set and all previous releases as the training set.

For each classifier used, a .csv file is generated containing for each combination of techniques used the values related to prediction:

Metric	Description
PRECISION	How many times you have correctly classified an instance as positive
RECALL	How many of the positives you have been able to classify
AUC	Area subtended by the ROC curve
KAPPA	How many times you have been more accurate than a dummy classifier
ACCURACY	Percentage of all right estimates out of all estimates
TRUE-NEGATIVE	Predicted negative and actually negative
TRUE-POSITIVE	Predicted positive and actually positive
FALSE-POSITIVE	Predicted positive and actually negative
FALSE-NEGATIVE	Predicted negative and actually positive

Configuration File

The configuration.properties file was used for configuration. The parameters are:

project	Name of the project to be analyzed
use_local	Path of the cloned project locally
project_path	Path of the cloned project locally
coldstart_project	Set to 'true' to use an apache repository already cloned locally or set to 'false' to automatically clone the repository
threshold	Minimum number of commits to make proportion increment otherwise make cold start
aproximate	If set to 'true' it approximates the Injected Version calculated via proportion by excess, otherwise it approximates by default
get_test_classes	If set to 'true' consider the project test files

Project Vulnerabilities

The project was analyzed via the Sonar Cloud platform ( 🔗 analysis ), showing the presence of 0 code smells, 0 bugs and 0 vulnerabilities.

Presentation 🇮🇹

An expository presentation of the project has been created in Italian language.