uORB_tests: add uORB::SubscriptionMultiArray tests

[dagar]

No description provided.

[dagar]

@bkueng what did you hit in #16092?

[dagar]

This also uncovers a small bug with uORB::DeviceNode::get_initial_generation() and topics that were previously published then unadvertised. @shawnfeng0 FYI

Possibly already resolved by #16084.

[shawnfeng0]

It may be because test_single() has used orb_test instance 0 to publish data once, but then you subscribe in test_SubscriptionMulti(), orb_test instance 0 has a piece of data waiting to be read, but it has not been read yet, then there is a new publication, resulting in the instance 0 produced two unread data, so the test failed.

[shawnfeng0]

That should be it. I turned off test_single() first and did the test, and it passed.

I see that in order to maintain independence, each unit test basically uses a different message. Maybe a new one should be used here. Or use copy several times in advance to read all the messages before executing the test.

[dagar]

That should be it. I turned off test_single() first and did the test, and it passed.

I see that in order to maintain independence, each unit test basically uses a different message. Maybe a new one should be used here. Or use copy several times in advance to read all the messages before executing the test.

That's all true, but I still don't see why this isn't technically a real bug. The first subscription subscribes and reports the last generation as 1 (should be 2 from the earlier test_single()), then after it does the first copy the generation jumps to 3 (actual generation + 1).

[shawnfeng0]

Oh, test_single() was published twice, orb_advertise(ORB_ID(orb_test), &t) and orb_publish(ORB_ID(orb_test), ptopic, &t).

Then test_SubscriptionMulti() was published once and it became 3, and that's it.

The generation is the total number of publishes, but it is used as an index starting from 0 inside the uORB::DeviceNode. When I first saw it, I found it a little strange, this is a small trick. But from the outside, it is the normal total number of releases.

[shawnfeng0]

The test failed because of this:

PX4-Autopilot/src/modules/uORB/uORBDeviceNode.cpp

Line 552 in 99833eb

unsigned generation = _generation.load() - (_data_valid ? 1 : 0);

If there is valid data before, we will reduce the initial value by 1 to get the latest data. Then it was released again in test_SubscriptionMulti(), our _last_generation = generation-2. In addition, there is no queue for the orb_test topic, so it was skipped twice when copying.

My modification in #16084 cannot fix this problem.

For this problem, we should use a copy before the start of this test case to point _last_generation to the latest data. Or it should be tested on a new topic.