I have several pieces of software in ROS, all of which are pretty thoroughly unit and integration tested. However, there's one aspect of testing within ROS that I have yet to accomplish with much success: Verifying publications.

Note that I'm still stuck in the dark ages of Fuerte, but I believe my question is relevant to more recent releases as well.

I currently have several tests within rostest, which looks like this in my CMakeLists.txt:

    rosbuild_add_executable(my_tests EXCLUDE_FROM_ALL ${MY_TESTS})
    rosbuild_add_gtest_build_flags(my_tests)
    rosbuild_add_rostest(${PROJECT_SOURCE_DIR}/test/node/node.test)

Running this with `make test` also runs a roscore (since node.test is a launch file), so these tests can make use of classes that require a connection to roscore. I'd like to use this to verify publications, but it's not working reliably, which makes me think I'm doing something wrong. As an example:

class TestSubscriber
{
    public:
        TestSubscriber() : receivedMessage(false) {}

        void callback(const std_msgs::Int32ConstPtr &newMessage)
        {
            receivedMessage = true;
            message = newMessage;
        }

        bool receivedMessage;
        std_msgs::Int32ConstPtr message;
}

TEST(MyTest, TestPublication)
{
    // This class is the one that will publish on the "output" topic. It will create a node handle
    // and advertise in its constructor.
    MyPublishingClass publishingClass;

    ros::NodeHandle nodeHandle;

    TestSubscriber subscriber;
    ros::Subscriber rosSubscriber = nodeHandle.subscribe("output", 1, &TestSubscriber::callback, &subscriber);
    ASSERT_EQ(1, rosSubscriber.getNumPublishers()); // This passes

    publishingClass.publishMessage();

    ros::spinOnce(); // Spin so that publication can get to subscription

    EXPECT_TRUE(subscriber.receivedMessage); // This may or may not be true
}


Depending on how the MyPublishingClass is written, this test may pass or fail due to that last line. To give a specific example, I have a class that reads through a bagfile and publishes specific topics. If that class publishes topics with:

    publisher.publish(rosBagViewIterator->instantiate<std_msgs::Int32>());

the test passes. If instead the class publishes with:

    publisher.publish(*rosBagViewIterator);

the test fails (even though the node still runs normally outside of the test, so I know that publication is actually happening). I have more examples of similar weirdness, but I'm hoping this is enough to describe my problem.

So, down to my questions:

1) Why is this so finicky? I realize that it's basically a node subscribing to its own publication, but that shouldn't be a problem. Should it?
2) Am I doing this wrong? Is there a better way to test ROS publications? Note that the class in question is _not_ a complete ROS node, and cannot be tested as such.

Thank you for your help!

--------------------
Kyle Fazzari
Computer Engineer
kyle.fazzari@navy.mil