Hey folks, I have a PR up that changes which commit we build in our PRs. Currently, our PRs are building a synthetic "merge" commit which does not actually exist in the repo. This is done by taking the base branch and merging the PR into it. The benefit of this approach is that you will see the results of your PR after a merge, so it's more realistic.
However, the problem with this approach is that it frequently makes the build cache useless. Since we often have many commits on trunk which have not yet been built (and cached), when these are pulled into the PR via the merge ref it will cause cache misses. I have a simple patch which changes our PRs to build the HEAD of the PR rather than the merge ref. https://github.com/apache/kafka/pull/18449 I have set the "Compile and Validate" step to continue building the merge ref. This will let PR authors know about integration issues early on. Let me know what you think. -David A
