How kmura88/sbfl_command works
Project Type and Purpose
• This is an open-source CLI tool for Java Suspicious Behavior Fault Localization (SBFL).
• It orchestrates Java test execution (JUnit3/4), collects coverage via JaCoCo, converts it to XML, then runs Python-based analysis to compute fault‐localization metrics (e.g., Ochiai, Tarantula).
High-Level Architecture Components
a. CLI Wrapper Scripts (.cmd)
• SBFL-JUnit3.cmd, SBFL-JUnit4.cmd, xml_maker_junit3.cmd, xml_maker_junit4.cmd
• Accept user arguments (test class, target class), set up environment, invoke downstream steps.
b. Java Execution Layer
• Classes in classes/ (compiled output) and source in src/:
• MethodParserAndRunner(.java) & its JUnit3 variant: extract @Test methods from test classes
• RunSingleTest(.java) & its JUnit3 variant: invoke individual test methods via JUnitCore
• Relies on JUnit jars (lib/junit/…) on the classpath.
c. Coverage Collector (JaCoCo)
• Agent jars in lib/jacoco/ instrument bytecode at runtime
• Generates .exec files (one per test‐method).
d. XML Generator
• XML conversion driven by xml_maker_*.cmd scripts
• Processes each .exec into an XML report describing line coverage & test pass/fail.
e. Python Analysis Layer (in lib/)
• XmlAnalyzer.py – filters out SBFL tool classes + test classes, transforms XML to boolean coverage per line
• LineExecutionCounter.py – tallies execution counts from XML reports
• SBFL_base.py – computes nf (not failed hits), ef (executed failed hits), np, ep, writes raw results to CSV
• SBFL_Ochiai.py & SBFL_Tarantula.py – extend SBFL_base with specific suspiciousness formulas
f. File System / Data Stores
• Input: classes/ (compiled .class files), src/ (source .java for recompile)
• Intermediate: per‐test .exec files, per‐test XML coverage files
• Output: CSV ranking lines by suspiciousness
Component Interactions & Data Flow
1) User runs SBFL-JUnit4.cmd testClass targetClass
2) CLI script checks args, ensures classes/ and src/ are populated
3) MethodParserAndRunner parses test class for method names → invokes xml_maker_junit4.cmd per method
4) xml_maker:
• Launches java -javaagent:jacocoagent.jar … RunSingleTest testClass#method
• Produces .exec via JaCoCo
• Runs JaCoCo CLI to emit XML per .exec
5) For each XML: Python XmlAnalyzer filters & converts to boolean coverage → LineExecutionCounter tallies
6) SBFL_base collects nf, ef, np, ep → writes intermediate CSV
7) SBFL_Ochiai and SBFL_Tarantula compute suspiciousness → augment final CSV output
Architectural Patterns & Design Principles
• Pipeline architecture: sequential transformation of data (test definitions → exec → XML → counts → metrics).
• Separation of concerns:
• CLI orchestration (.cmd)
• Java for test execution & coverage capture
• Python for data parsing & metric computation
• Extensibility points: adding new SBFL metrics by subclassing SBFL_base, supporting JUnit5 by adding new runner classes.
Detailed Elements to Capture
• .cmd scripts as entry points
• Distinct Java classes for test parsing vs. test execution
• JaCoCo agent injection and its output artifacts
• XML conversion step and filtering logic (ignoring tool classes/tests)
• Python modules with clear inheritance from SBFL_base
• Final CSV generation