Bug Triage

Open source software projects typically have a bug repository that allows both developers and users to post problems encountered with the software, suggest possible enhancements, and comment upon existing bug reports. Popular open source projects receive lots of bug reports.

Since more bugs are reported than can be easily handled, each bug must be triaged to determine if it describes a meaningful new problem or enhancement, and if it does, it must be assigned to an appropriate developer for further handling.

We are investigating approaches to help automate two parts of the bug triage process:

• Bug Report Assignment (a.k.a. Sibyl)
• Duplicate Detection

One step in bug triaging is assigning a developer to a newly received report. We are developing an approach for semi-automating this assignment. Our approach uses a machine-learning algorithm that is applied to the information about how bugs have been assigned in the past to recommend developers to whom a new bug should be assigned.

Sibyl is a web-service implementation that adds recommendations to a Bugzilla bug page and is being used to evaluate our approach. Video tutorials showing how to set up Sibyl and how to use the Sibyl Triage Advisor are available. Currently Sibyl provides recommendations for the following projects:

• Eclipse Platform
• Firefox Browser

E-mail John Anvik to find out if Sibyl could be used for your project. As a guideline, the project should:

1. Use Bugzilla (V.2.16 or higher) for tracking bug reports where the default layout for bug reports has not been altered significantly.
2. Contain more than 20 developers who actively fix bugs.
3. Had 500+ reports marked as FIXED in the last 8 months.

Triagers need to identify which new bug reports are duplicates of previous reports. In large projects where there are thousands of reports to search through, identifying duplicate bugs can be a daunting task.

We have developed an approach to suggest to a triager possible duplicates of a new report. Our approach is similar to the techniques used by event detection systems from Topic Detection and Tracking (TDT) research that identify whether a news story describes a new, previously unseen event or a old news event.

We are currently conducting a study to determine how beneficial the duplicate detection tool is for bug triagers. We are looking at what effects the tool would have on a triager's duplicate detection accuracy and their workload.

Future plans include implementing the tool as a Firefox extension.

If you would like to be notified when the tool becomes available, email Lyndon Hiew.