Closed dabonnie closed 4 years ago
- Consider pinning navigation a commit ID instead of a branch so that we don't get broken because nav is broken
Good idea.
- Did you choose a large package on purpose? I'd suggest having a big one and a trivial one so that we can easily repro simple issues (if we can't bundle ament_cmake something is wrong, if nav can't be bundled... lots fo things go wrong)
Yes, see 4.
- How flaky is this? I'd suggest running that at a high frequency on a fork for a while to test this out?
That's a good question: hourly would help here.
- You should be using the Docker image owned by ros-tooling to avoid APT repo flakiness.
Actually, apt repo flakiness is something I wanted to test for given https://github.com/colcon/colcon-bundle/issues/178 and a patch that potentially addresses the problem.
would it make sense to have another test for apt flakiness? That way this nightly workflow is not testing for multiple variables.
would it make sense to have another test for apt flakiness? That way this nightly workflow is not testing for multiple variables.
Possibly, though I'd like to have a full workflow test as well.
Changes look good to me. I'd second Thomas' suggestion of testing on one trivial package (ament_cmake
maybe?) and one big package (navigation
seems fine). This way can distinguish bundle failures vs. other failures brought by big packages.
This is ready for re-review @thomas-moulard @prajakta-gokhale @zmichaels11.
If review passes I will open a separate PR in order to define the vcs repo file and have this PR point to that in order to merge with green CI.
Also currently 26 successful runs via actions on a fork (no failures) since last night run at every 30 minutes. Here's the link: https://github.com/dabonnie/colcon-bundle/actions?query=event%3Aschedule
Superseded by https://github.com/colcon/colcon-bundle/pull/191
Build and bundle a ROS1 package (navigation) using the latest released colcon-bundle (via pip).
Signed-off-by: Devin Bonnie dbbonnie@amazon.com