• Linux
• Linux-aarch64
• macOS
• Windows

• Linux
• Linux-aarch64
• macOS
• Windows

It was a symbol visibility issue for the test executable to link the qos utility functions. Added the explicit yaml dependency while I was poking around

Sanity re-test for windows - the failures seem really unrelated and are potentially just flaky
Double sanity check on master without this change

@Karsten1987 the buildfarm is showing the exact same test failures on windows on master as with this PR. do you think we're good to merge?

@Karsten1987 this particular test is failing on the nightly build as well https://ci.ros2.org/view/nightly/job/nightly_win_deb/1573/testReport/junit/rosbag2_tests/RecordFixture/record_end_to_end_test_with_zstd_file_compression_compresses_files/ - in order to not block this feature due to unrelated issues, I'm going to merge now

while debugging #329 I've noticed the following test error with cyclonedds:

[1.348s] 7: /home/karsten/ros2_ws/src/ros2/rosbag2/rosbag2_transport/test/rosbag2_transport/test_record.cpp:99: Failure
[1.349s] 7: Value of: serialized_profiles
[1.349s] 7: Expected: contains regular expression "- history: 3\n  depth: 0\n  reliability: 1\n  durability: 2\n  deadline:\n    sec: 2147483647\n    nsec: 4294967295\n  lifespan:\n    sec: 2147483647\n    nsec: 4294967295\n  liveliness: 1\n  liveliness_lease_duration:\n    sec: 2147483647\n    nsec: 4294967295\n  avoid_ros_namespace_conventions: false"
[1.349s] 7:   Actual: "- history: 1\n  depth: 10\n  reliability: 1\n  durability: 2\n  deadline:\n    sec: 9223372036\n    nsec: 854775807\n  lifespan:\n    sec: 9223372036\n    nsec: 854775807\n  liveliness: 1\n  liveliness_lease_duration:\n    sec: 9223372036\n    nsec: 854775807\n  avoid_ros_namespace_conventions: true"
[1.352s] 7: [  FAILED  ] RecordIntegrationTestFixture.qos_is_stored_in_metadata (172 ms)

Thanks for pointing that out! I think we should call this a bug in rmw_cyclone.

See how the different values are the first two, around history.

• History/depth policy have no bearing on compatibility, so it is not useful to know what the endpoint uses
• The other RMW implementations return RMW_QOS_HISTORY_POLICY_UNKNOWN and depth=0, which makes all endpoint info consistent, which makes it much easier to respect the above point

EDIT: I now see that the "infinity" values for the rmw_time_t types are also different. This is problematic in a different way. FastRTPS and Connext were both returning the same values for infinity, but Cyclone seems to be using something different.

The values I was expecting here were sec = 0x7FFFFFFF (signed 32 bit int max) and nsec = 0xFFFFFFFF (unsigned 32 bit int max, or -1?).

The values Cyclone is returning are sec = 9223372036 == 0x225C17D04 and nsec = 854775807 == 0x32F2D7FF which my brain sees as totally random.

do you mind reporting this on https://github.com/ros2/rmw_cyclonedds? I feel you have more insights on QoS than me to give a meaningful issue report ;-)

However, just for my understanding: If history/depth have no influence, why are we testing for it then? We rely on every rmw implementation to adhere to some sort of default then. Are these defaults documented?

I think it's reasonable to remove the check for it in the test you saw failing. I'm worried about cross-rmw compatibility, seeing these values, though.

EDIT: see previous comment for more analysis on the values