我们的系统应该满足这四个requirements.ppt

Analyzing Individual Performance of Source Code Review Using Reviewers Eye MovementHidetakeHidetake UwanoUwano,Masahide Nakamura,Masahide Nakamura,AkitoAkito MondenMonden and Ken-and Ken-ichiichi Matsumoto MatsumotoNara Institute of Science and TechnologyNara Institute of Science and TechnologyETRA 2006ETRA 2006IntroductionCapturing Eye Movement in Code ReviewExperimental Evaluation of Individual Performance of Code Review Using Eye MovementDiscussionConclusionMy ThoughtIntroductionSource code review is intended to find and fix defects overlooked in early development phases.The long-term goal is to establish an efficient method that allows the reviewer to find as many defects as possible.Several methodologies that can be used for the code review have been proposed,but there has been no significant conclusion on which review method is the best.The reason is that the performance of individual subjects is more dominant than the review method itself.To characterize the reviewers performance in an objective way,this paper proposes to use eye movements of the reviewer.They developed a software application,“Crescent(Code Review Evaluation System by Capturing Eye movemeNT)”,to measure and record the eye movements during the code review.Capturing Eye Movement in Code ReviewRequirementsR1:Line-wise tracking of eye movementsR1:Line-wise tracking of eye movementsR2:Identification of reviewers focusR2:Identification of reviewers focusR3:Record of time-sequenced linesR3:Record of time-sequenced linesR4:Analysis supportsR4:Analysis supportsIntegrated Measuring Environment1.Computes the relative coordinates of the eye mark in the textbox.2.Converts each relative coordinate into a line number.1.Monitors all events of the slider bar2.Adapts the correspondence to maintain the consistency of the line numberSummarize the focused line as a time-sequenced data.Play back and visualize the recorded line informationSend a reset signal to the eye mark trackerPops up a main textboxdisplaying a source code to be reviewedExperimental Evaluation of Individual Performance of Code Review Using Eye MovementPreliminariesFive graduate students participated in the Five graduate students participated in the experiment as the reviewers.experiment as the reviewers.They have prepared 6 small-scale programs They have prepared 6 small-scale programs written in the C language.written in the C language.Programs reviewed in the experimentPrograms reviewed in the experimentThe fixation criteria as the area of 30 pixels in The fixation criteria as the area of 30 pixels in diameter where the eye mark stays more than diameter where the eye mark stays more than 50 ms.50 ms.Task of code reviewCalibrate the eye tracker so that that eye Calibrate the eye tracker so that that eye movement of the subject are logged correctly.movement of the subject are logged correctly.Explain the specification of the program to the Explain the specification of the program to the subject verbally.subject verbally.Synchronizing the subject to start the code Synchronizing the subject to start the code review to find defect,start the capture of eye review to find defect,start the capture of eye movements and code scrolling.movements and code scrolling.Suspend the review task when the subject tells Suspend the review task when the subject tells he/she find defect.Then,ask the subject to he/she find defect.Then,ask the subject to explain the defect verbally.explain the defect verbally.Finish the code review task if the detected Finish the code review task if the detected defect is correct.Otherwise,resume the task defect is correct.Otherwise,resume the task going back to the step 3.The review task is going back to the step 3.The review task is continued until the subject successfully finds the continued until the subject successfully finds the defect,or the total time for the review exceeds 5 defect,or the total time for the review exceeds 5 minutes.minutes.Analyzing ResultCharacterizing Review Performance by Scan PatternDiscussionEye Movements and Reviewers ThoughtTwo kinds of interviewTwo kinds of interviewFirst interview:Show the source code and First interview:Show the source code and asked what the subject had been thinking in the asked what the subject had been thinking in the code review.code review.Second review:Show the eye movements and Second review:Show the eye movements and the source code,and asked the same questions.the source code,and asked the same questions.It seems that the record of the eye movements It seems that the record of the eye movements reminded the subjects of their thought well.reminded the subjects of their thought well.Other Reading PatternsRetrace Declaration PatternRetrace Reference PatternConclusionThey developed an integrated measuring environment of eye movements in code review.Conducted an experiment and found a particular reading pattern,called scan.It was shown that the reviewers taking sufficient time for scanning the code tend to detect defects efficiently.My Thought我們的系統應該滿足這四個requirements系統設計的架構