henry199898/drone

DRONE

Description

DRONE( Detect and Repair of dOcumentatioN dEfects ) is integrated into a framework, which can automatically detect and repair defects of API documents by leveraging techniques from program analysis, natural language processing and constraint solving.

Our approach includes several parts as follows:

• Leverage static analysis techniques for program comprehension
• Leverage domain specific, pattern based natural language processing (NLP) techniques for document comprehension
• The above two steps' analysis results are presented in the form of first-order logic (FOL) formulae
• The generated FOL formulae are then fed into the SMT solver Z3 to detect the potential defects in case of inconsistency
• In the end, a template based patch will be recommended to users, suggesting how to repair the defects

Our approach mainly handles four types of document defects, which include null not allowed, null allowed, range limitation and type restriction.

We summarize more than 60 heuristics on the typical descriptions of API usage constraints in documents, which could be reused across different projects.

DRONE generates some templates by analyzing both extracted code constraints and document patterns for each of the four constraints types.

Configuration

• step 1. Import the project DRONE into Eclipse IDE;
• step 2. The two files libz3.dll and libz3java.dll need to be put under the specified folder: C:/Windows/System32/ ;
• step 3. Please make sure that vcredist_x64.exe has been installed (Our tool uses the SMT solver Z3, so step 2 and 3 need to be executed.);
• Make sure that the dependent Libraries of the two files libz3.dll and libz3java.dll exist( You can download Dependency Walker from the website: http://www.dependencywalker.com/ to check whether the dependent Libraries are complete).
• Note that the version of the JDK we use in our current implementation is Java SE Development Kit 8u144 and the verison of eclipse is Luna Service Release 2 (4.4.2) .

(Current release only supports Windows OS)

Usage

• step 1. Project should run as Eclipse Application;
• step 2. The tool contains two views: DRONE-Detect View and DRONE-Repair View (Windows->Show view->Other->sample category-> DRONE-Detect View and DRONE-Repair View);
• step 3. You need to put the projects to be detected under the folder runtime-EclipseApplication which has the same root dictionary with the workspace of Eclipse.
• step 4. After clicking the corresponding buttons in the DRONE-Detect View step by step , you can finally find a folder which stores the final results and the DRONE-Repair View will be presented to you. Under the folder ./CellDocViaInvokations/ , you can find a file checklog.csv.
• step 5. Select the file checklog.csv as the input and click start button and then you will see the recommendation result.

Explanation of the file formats of Experiment Results

• The corresponding experiment results are all under the folder of Replication
• The experiment results from 1 to 2 are all recorded in the files named of experiment1.xls, experiment2_part1.xls and experiment2_part2.xls respectively. Note that experiment 2 has two parts.
• With respect to the file experiment1.xls, column 1 records the corresponding links to the version where the the defects exist, column 5 and 6 record the corresponding FOLs generated from directives and source codes, respectively. Column 7 records the SMT-LIB2 formats from corresponding directives. Column 8 records the SMT-LIB2 formats from corresponding source codes. Column 9 records the analysis results of our tool "DRONE" and column 10 records the manual classification.
• The layout of the files experiment2_part1.xls and experiment2_part2.xls is the same as that of the file experiment1.xls.
• The files under the folder experiment3 record the recommendations of the corresponding detected defects of experiment1 and experiment2.
• The files under the folder AssociatedMethodBody record the 3915 detected defects(1689+1605+621), which are recorded in the file inconsistent, and the associated method bodies of the 3915 defects, which are recorded in the files named of APISource_experiment1, APISource_experiment2_part1 and APISource_experiment2_part2.
• The files under the folder StatisticalOfVerbsAndSubjects contain the statistics of the POS process over the 34874 tags: how many with subject? with verbs? what are the most commons verbs?