Switch to custom lttng-ctl Python bindings

[christophebedard]

Requires ros2/ci#725 for CI

This replaces the lttng-ctl Python bindings (python3-lttng), on which the ros2 trace command and Trace launch action rely, with a new implementation.

The new lttng-ctl Python bindings implementation, lttngpy, uses pybind11 to bind C++ code to Python. The bindings do not cover the whole trace control library (liblttng-ctl, from the liblttng-ctl-dev package on Ubuntu), but they cover whatever is needed by tracetools_trace/ros2trace. A lot of lttngpy is just exposing C functions and constants from lttng-ctl (which usually start with lttng_* or LTTNG_*) to Python. However, lttngpy does wrap some lttng-ctl functions to provide a more Pythonic interface, e.g., Optional[...] arguments and Union[...] return types.

I replaced the calls to the current Python bindings (lttng) in tracetools_trace (and some other locations) with calls to lttngpy. It is (almost) a simple drop-in replacement; users should not notice any difference. In fact, this even improves the usefulness of ros2 trace/Trace, since they now support enabling perf counters, e.g., ros2 trace -c perf:thread:task-clock. Indeed, the bindings available from python3-lttng do not completely cover liblttng-ctl. They are always lagging behind; I've fixed issues and contributed improvements before, but, since any new improvements are only going to be available from apt in the next Ubuntu LTS release, we cannot rely on this. Therefore, I think implementing our own bindings is a better solution.

Note that RHEL has liblttng-ctl version 2.12.11, while Ubuntu has 2.13.x, so I had to adapt some things to support both 2.12 and 2.13.

Note: The Sanity checks / binary (rolling) CI job will fail because ros-rolling-lttngpy is of course not yet available.

[christophebedard]

@iluetkeb as we discussed

[iluetkeb]

Awesome work, thanks :-)

[mjcarroll]

Overall the approach looks good. While it isn't zero-cost, I think writing your own bindings is the right call in this situation, especially when dealing with LTS durations.

I had one small question of something that jumped out at me as kind of odd, but overall LGTM

[christophebedard]

Testing everything above lttngpy with ros2/ci#725:

• Linux
• Linux-aarch64
• Windows

[christophebedard]

Oh, right, I need to install liblttng-ctl-dev in CI.

[christophebedard]

@mjcarroll I don't have access to ros2/ci. Could you push my PR branch to the repo? Then I can re-run the CI jobs above using that CI branch, or you can do it if you want.

[mjcarroll]

https://github.com/ros2/ci/tree/christophebedard/install-liblttng-ctl-dev

[christophebedard]

RHEL test job:

• Linux RHEL

[christophebedard]

The RHEL job fails due to another issue, see ros2/ci#725 (comment).

[christophebedard]

Let's try again:

Testing everything above lttngpy with ros2/ci#725:

• Linux
• Linux-aarch64
• Windows
• Linux RHEL

[christophebedard]

Looking good now, RHEL included.

@mjcarroll could you review the latest changes, i.e., the last commit?

[ros-discourse]

This pull request has been mentioned on ROS Discourse. There might be relevant details there:

https://discourse.ros.org/t/ros-2-jazzy-jalisco-released/37862/9