RV-Predict requires Java Runtime Environment 1.8 or higher.


Download the installer from the RV-Predict website and execute it with java -jar <installer> (unless your browser executes it automatically upon download), following all instructions. Remember to add the bin directory under the RV-Predict installation directory to your PATH environment variable.

Running RV-Predict

RV-Predict can be run both from the command line, as a drop in replacement for the java command, and as an agent, to ease integration with IDEs and build management tools like Maven.

On the command line

RV-Predict is invoked as follows:

    rv-predict [rv_predict_options] [--] [java_options] class [args...]
        (to predict races in a class)
or  rv-predict [rv_predict_options] [--] [java_options] -jar jarfile [args...]
        (to predict races in an executable jar file)

where [rv_predict_options] are RV-Predict options and [java_options] are normal Java options. Whenever it might cause confusion, the optional -- can be used as a terminator for the RV-Predict options.

As an agent

Assuming <rvPath> is the installation directory for RV-Predict, running RV-Predict as an agent along with your Java application simply requires adding the -javaagent:<rvPath>/lib/rv-predict.jar option and the -Xbootclasspath/a:<rvPath>/lib/rv-predict.jar option to your Java command line. In addition, we strongly recommend you to also add the -XX:hashCode=1 option; this significantly reduces the possibility of false positive due to identity hash code collision. Passing options to the agent can be done as standard for agents: using -javaagent:<rvPath>/lib/rv-predict.jar="<rv_predict_options>", where <rv_predict_options> are RV-Predict options.

The agent uses the z3 library as a constraint solver. Therefore, the z3 libraries need to be accessible from the library path. To ensure that for Linux/OSX systems (not necessary for Windows), please go to the <rvPath>/lib directory and source the setenv script:

<rvPath>/lib$ source setenv

Integration with Maven

For Maven-based projects which have tests, one can simply run mvn test, after modifying the individual project’s pom.xml to have an element similar to the following:


Replace ${surefire-version} with the exact surefire plugin version used by the project (e.g., 2.16).

Adding the -javaagent and -Xbootclasspath options are the only changes needed to an existing project and tests can still be run with mvn test. Again, the -XX:hashCode=1 option is optional but highly recommended.

Integration with IDEs

Generic instructions
options -javaagent, -Xbootclasspath and -XX:hashCode=1 (optional) need to be added to the VM options of your Run/Debug Configurations.
From the menu select Run -> Run Configurations -> (then you select the configuration that you are running) -> select Arguments tab -> enter -javaagent:<rvPath>/lib/rv-predict.jar -Xbootclasspath/a:<rvPath>/lib/rv-predict.jar into the VM arguments field.
IntelliJ IDEA
From the menu select Run -> Edit Configurations -> (then you select the configuration that you are running) -> enter -javaagent:<rvPath>/lib/rv-predict.jar -Xbootclasspath/a:<rvPath>/lib/rv-predict.jar into the VM options: field.

Tuning RV-Predict

The list of common options can be obtained by using the -h or --help option when invoking RV-Predict:

rv-predict -h

Usage: rv-predict [rv_predict_options] [--] [java_options] <java_command_line>
  Common options (use -h -v for a complete list):

      --offline          Run prediction offline

      --log              Record execution in given directory (no prediction)

      --predict          Run prediction on logs from given directory

      --include          Comma separated list of packages to include

      --exclude          Comma separated list of packages to exclude

      --window           Window size (must be >= 64)
                         Default: 1000

      --stacks           Record call stack events and compute stack traces
                         in race report

      --suppress         Suppress race reports on the fields that match
                         the given (comma-separated) list of regular

  -v, --verbose          Generate more verbose output

      --version          Print product version and exit

  -h, --help             Print help info


  • the --offline option tells RV-Predict to store the logged execution trace on disk and only run the prediction algorithm after the application terminates.
  • the --log <dir> option tells RV-Predict that the execution should be logged in the <dir> directory and that the prediction phase should be skipped.
  • the --predict <dir> option tells RV-Predict to skip the logging phase, using the logged trace from the <dir> directory to run the prediction algorithms. When using this option, specifying the java options and java command line are no longer necessary.
  • the --include option tells RV-Predict to include the given packages in instrumentation; this option takes precedence over the following --exclude option.
  • the --exclude option tells RV-Predict to exclude the given packages from instrumentation.
  • the --window <size> (default: 1000) option tells RV-Predict to find races between events with the largest distance of <size> in the logged trace. The larger the <size> is, the more races are expected to be detected, and the more time RV-Predict will take.
  • the --stacks option tells RV-Predict to record call stack events that can be used to compute stack traces in the race report.
  • the --suppress option tells RV-Predict to suppress race reports on the fields that match the given regular expression patterns; only used when the user is absolutely certain that the data race to be suppressed is benign.
  • -- can be used as a terminator for the RV-Predict options.

Advanced options

The complete list of RV-Predict options can be obtained by combining the -h and -v options:

rv-predict -h -v

As this list of advanced options is continuously evolving, we only list the more common ones here. Please feel free to contact us in case the explanations displayed by rv-predict -h -v are not sufficient:

  • the --profile option instructs RV-Predict to run in the profiling mode which does not perform any deep analysis. It is commonly used to estimate the number and distribution of events generated from the instrumented classes.
  • the --solver <solver> option instructs RV-Predict to use a different SMT solver for handling SMT queries.

Additionally, the RV_OPTS environment variable can be used to implicitly pass extra options to the Java Virtual Machine running RV-Predict. RV-Predict will pick up these Java options upon start-up. For example, RV_OPTS=-Xss32m -Xms1g -Xmx3g sets the thread stack size of the JVM to be 32MB, initial heap size to be 1G, and maximum heap size to be 3G. Of course, you can still explicitly pass -Xss32m as a Java option as described in the previous section.

Enhancing prediction power

By default, RV-Predict tries to keep a good balance between efficiency and prediction power. Nevertheless, while the default settings were engineered to work for most common cases, there might be cases where user input could improve the prediction process. We provide several options for advanced users to tune RV-Predict:

  1. Window size. For efficiency reasons, RV-Predict splits the execution trace into segments (called windows) of a specified size. The default window size is 1000; however, the user can alter this size using the --window option, with the intuition that a larger size provides better coverage, at the expense of increasing the analysis time.
  2. Excluding packages. To allow better control over the efficiency, RV-Predict provides the option --exclude to remove certain packages from logging. This option takes a list of package patterns prefixes separated by , and excludes from logging any class matched by one of the patterns. The patterns can use * to match any sequence of characters. Moreover, * is automatically assumed at the end of each pattern (to make sure inner classes are excluded together with their parent). Please note that excluding packages might affect precision, as events from non-logged packages might prevent certain race conditions from occurring. Note: in bash-like enviroments, the $ character must be escaped as it is used by the shell to introduce environment variables.
  3. Including packages. To give more flexibility to selecting which packages to include and exclude, RV-Predict also provides the --include option which is similar to the --exclude option (comma separated list of package patterns), but opposite in effect.

Problems running RV-Predict?

We list below some possible issues occurring when using RV-Predict and ways to address them. For any unlisted issue you might experience, please use the RV Support Center.

Program does not seem to terminate

The execution of the program takes too long when run using RV-Predict.
It could be due to the overhead required by RV-Predict analysis, or due to a deadlock condition triggered by the logged program.
You can stop the program at any time and run the prediction phase on the already logged trace using the --predict option with the directory in which the trace was logged (printed by RV-Predict when the logging was started).

Stack overflow error

I’m getting an unexpected Stack Overflow exception and a huge stack trace when running my program with RV-Predict.
The execution trace to be analyzed is collected by RV-Predict using a Java agent, which means that the call stack of the logging module adds on top of the call stack of the original application.
Try increasing the stack size of the logged program by passing the -Xss option to RV-Predict.