[CI Problem] unity/pr-head does not automatically rebase against main before CI run

[Luke Hutton]

unity/pr-head started failing due to a concurrent merge of both #16612 and #16563. This issue was fixed in #16649, however, PR's that haven't been rebased on the latest main are still failing.

Other CI runs e.g. cpu/pr-head, gpu/pr-head, etc automatically rebase the PR before a CI run. This ensures the PR's are tested against the latest main, rather than a previous head of main. unity/pr-head doesn't seem to do this which can result in PR's failing even after triggering a rerun of CI:

• https://discuss.tvm.apache.org/t/is-there-an-anomaly-in-ci-unity-pr-head/16446/3
• #16652 (comment)

The behaviour of unity/pr-head should be consistent with cpu/pr-head, gpu/pr-head, etc. Taking a look at the jenkinsfiles, it seems unity/pr-head is missing a call to merge_with_main()(link) when a PR branch is detected.

Originally posted here, creating an issue for better visibility.

[Zheng-Bicheng]

@lhutton1, thank you for addressing my concern. I've noticed some anomalies in TVM's CI. Firstly, there seems to be a strange error occurring during the Build phase of my PR in CI tvm-arm (I haven't made any modifications to the C++ code). Secondly, CI tvm-gpu has been stuck in a queued state for what appears to be a day now, with no changes thus far.

[Yong Wu]

Tried to add the merge_with_main in unity #16661

[Luke Hutton]

Closing since #16661 was merged