When you're writing test suites for your code, it can be very useful to use a tool to analyze the code coverage of your tests. While people can get a little obsessive about coverage percents, there's certainly no denying that it's helpful to know how much of your code is actually run when testing, and also being able to look down into the specifics of what is covered.

With Java, one of the preeminent tools for this is JaCoCo, a venerable open-source library that you can integrate with your test suites to give reports of your coverage. In a normal project, such as a build run via Maven, you can use the Maven plugin in tandem with the Maven Surefire and Failsafe plugins. However, things get more complicated if the code you're actually testing isn't in the Surefire JVM, but rather inside a container.

That's exactly the situation I have with the integration-test suite of the XPages Jakarta EE project, where it creates a Docker container with the current build of the project deployed as OSGi plugins, and then executes HTTP calls against OSGi bundles and NSFs. I figured this was a solvable problem, so I set out doing so.

I first came across this blog post, which describes the general idea well, but unfortunately references Gists that seem to no longer exist. Still, it gave me a good starting point.

Installing JaCoCo in Domino

The first thing I had to do was to get the JaCoCo Java agent into the container. I added it as a Maven dependency to the IT suite project:

1
2
3
4
5
6

<dependency>
	<groupId>org.jacoco</groupId>
	<artifactId>org.jacoco.agent</artifactId>
	<version>0.8.11</version>
	<scope>test</scope>
</dependency>

Conveniently, this dependency is itself a wrapper for the agent JAR and comes with a convenience method for accessing the JAR data. I used that to read it into memory and send it to the Docker runtime during the container build:

1
2
3
4
5

byte[] agentData;
try(InputStream is = AgentJar.getResourceAsStream()) {
	agentData = IOUtils.toByteArray(is);
}
withFileFromTransferable("staging/jacoco.jar", Transferable.of(agentData)); //$NON-NLS-1$

The use of Transferable here allows me to keep the process independent of whether Docker is running locally or remote - I run remotely almost all the time nowadays, due to Domino's continued lack of an ARM port.

With the file in place, I modified my Dockerfile to copy it to a known location in the container:

1
2

COPY --chown=notes:notes staging/jacoco.jar /local/
COPY --chown=notes:notes staging/JavaOptionsFile.txt /local/

The JavaOptionsFile.txt was already there for another ARM-related reason, but it's important to note for the next step. This sort of file is how you enable JaCoCo in the Domino JVM: I set JavaUserOptionsFile=/local/JavaOptionsFile.txt and it'll read its rules from there. Following the instructions, I added -javaagent:/local/jacoco.jar=output=file,destfile=/tmp/jacoco.exec on its own line in this file. This causes JaCoCo to be automatically loaded with the HTTP JVM and to store its report in the named file on shutdown.

Reading the Data

That said, this didn't work immediately. The file "/tmp/jacoco.exec" was created properly inside the container, so the agent was running, but the file content was always zero bytes. I realized that this was due to the merciless way in which the container is killed by my test suite: there's no proper shutdown step, and so JaCoCo's shutdown hook never fires.

Fortunately, writing to a file isn't the only way JaCoCo can do its reporting - you can also have it open up a TCP port to connect to and read. So I changed the Java option line to:

1

-javaagent:/local/jacoco.jar=output=tcpserver,address=*,port=6300

I modified the withExposedPorts(...) call inside the class that builds my Testcontainers container to also include 6300, and then used getMappedPort(6300) to identify the actual randomized port mapped by Docker.

The remaining task was to figure out the little protocol used by JaCoCo to signal that it should collect and return its data. I get the impression that it's not too complicated, but I still figured it'd be best to use an existing implementation. I found jacocotogo, a Maven plugin that reads the data, and it looked promising. However, it had two problems: being a Maven plugin, it came with a bunch of transitive dependencies I didn't want, and it's also 11 years old and thus a bit out of date.

I ended up forking the main utility class, trimming out the parts I didn't need (like JMX), switching it to NIO, and going from there.

Using the Data

With that all in place, a test run will end up with a file named "jacoco.exec" inside the "target" directory. Using this file varies by IDE, but, in Eclipse, you can install the EclEmma tool, open the "Coverage" view, right-click in the table area, and choose "Import Session...". That will let you locate the file and then choose the projects from your workspace that you're looking to analyze.

When I did that, I got my results:

This is surprisingly good for the project, especially when you consider how large chunks of the red bars are things like the servlet wrapper package, which includes a lot of delegating code that is obligatory to match the interface but is not likely to be actually used in practice.

While this is currently the only project where I've needed to do this, it'll certainly be good to keep these techniques in mind. The TCP port thing in particular should be handy in future edge cases even without the Docker part.