Check our the Online Demo.
Bixie is a static analysis tool that detects inconsistencies in Java bytecode. An inconsistency occurs when code must throw an exception or is unreachable because because of assumptions made by other statements.
if (operator == null) {
throw new SemanticException("Operator " + operator.getName());
}
In this example from Hive, operator.getName()
is inconsistent with the conditional
operator == null
.
In this example from Cassandra:
public Boolean generate() {
return identityDistribution.next() % 1 == 0;
}
There is an inconsistency in the bytecode because the expression identityDistribution.next() % 1 == 0
appears as a conditional choice in the bytecode, and one case is unreachable because identityDistribution.next() % 1
returns a constant value.
Bixie uses gradle to build:
git clone https://github.com/SRI-CSL/bixie.git
cd bixie
./gradlew shadowJar
To check if everything is working, run Bixie on itself:
cd build/libs/
java -jar bixie.jar -j ../classes/main/
For your convenience, there is a Python script (runner/runner.py) that will automatically determine what classpaths to use and where class files are generated, and then invoke Bixie. It works for projects built with ant, maven, or gradle. For example:
cd <path-to-project>
mvn clean
python <path-to-bixie>/runner/runner.py -- mvn compile
This command will execute the maven build process for the project and scrape its output for instances where javac was called, then feed that information to Bixie. Be sure to clean before building, as the tool can only detect files that were actually compiled while it is observing.
Bixie is not sound. Many Java features, such as concurrency and reflection, are not handled by Bixie and may result in false alarms. Bixie also sometimes detects inconsistencies in the bytecode that have no corresponding inconsistency in the source code. For example, conditional choices with conjunctions in the condition sometimes raise false alarms.