JDCallgraph - Dynamic call graph generation for Java. Uses ByteBuddy or Javassist to instrument the target application. Can also create a data dependency graph to show field access between methods. Support various output formats like csv, dot or a matrix.

Download the jdcallgraph.jar or build it using mvn install. You have to invoke java with the -javaagent:/path/to/jdcallgraph.jar=/path/to/config.ini. A few examples:

java -javaagent:/path/to/jdcallgraph.jar=/path/to/config.ini example.Main

or

java -javaagent:/path/to/jdcallgraph.jar=/path/to/config.ini -jar target.jar

Simply invoke Maven with the -DargLine=-javaagent:/path/to/jdcallgraph.jar=/path/to/config.ini parameter. There is a wrapper script you can use like normal Maven:

./mvn.sh test -f /path/to/pom.xml

I did not succeed starting it with the ANT_OPTS environment variable. Instead you can edit your build.xml: Change the lines

<junit fork="no">
  ...
</junit>

To

<junit fork="yes" forkmode="once">
  <jvmarg value="-javaagent:/path/to/jdcallgraph.jar=/path/to/config.ini" />
  ...
</junit>

You can pass the path to a configuration file in the following way:

java -javaagent:jdcallgraph=/path/to/config.ini -jar target.jar

The config file has to contain key-value-pairs in a syntax similar to YAML. Whitespace as well as lines starting with # are ignored:

# this is a comment
key1: false
# a list of values
key2: 1, 2, 3

You can check out the configurations in the example folder. Options not present in your config take the value from defaults.ini The following options are supported:

Specify an output directory. Either absolute or relative path to a folder. JDCallgraph will put all the resulting graphs and log files into this folder.

Specify a log level. From 6 - TRACE (the most verbose) to 0 - OFF (no logging).

Switch logging to console on or off. When set to true it will not only log to the logfiles but also print log statements to the console. The output is mixed with the output of the original program.

Whether the resulting graphs should contain duplicate edges. Note that JDCallgraph has to keep a list of known edges which might lead to memory issues with large graphs.

Specify the criteria that is used to group the graphs. It will output one graph per category. Valid options are:

• THREAD: The default. One graph per thread of the observed program.
• ENTRY: A new graph for each entry method. An entry method is every interesting method with an empty stacktrace

Specify one or multiple output formats. Multiple formats are separated by comma: DOT,TRACE. Valid options:

• DOT: Write the graph to a .dot file. You can transform it to a png using dot -Tpng file.dot -o file.png
• MATRIX: Output the coverage matrix to a file called matrix.csv. The file will contain one row per entry method and one column per covered method. For each entry-method-pair it either outputs 1, when it was covered, or 0 otherwise.
• TRACE: Write a file called trace.csv. It contains one row per entry method and a list of methods that come after behind (Typically a list of methods covered per test).
• COVERAGE: Write a coverage.csv file. The opposite of TRACE. For each method it lists the tests this method was covered in.
• DATA: Also compute data dependency graphs. This will output them as .dot graph.

Decide which backend to use for the instrumentation. Since v0.2 ByteBuddy is the default (javassist: false). You can change this by setting this option to true. Note that Javassist is not compatible with Java 9, also the ByteBuddy backend is more tested.