Continuous Integration (CI) is great: basically you have a system that ensures that every time a change is committed your full test suite is run, and any failures are reported back. In our case, we have tests that take around 40 minutes to run (because we have lots of tests that need to create data and then ensure that things work based on that data), so being able to have that happen while you continue work is really nice. On top of this, we use Codecov to check that coverage does not decrease on a given commit.
Our general workflow is that we create a branch (in Mercurial, branches are different to git branches, they are long-lived, and all code within a branch retains that reference, meaning it’s easy to associate code with Jira issues), and create a Pull Request when the code is ready for review. At any one time there may be several Pull Requests open, that may or may not affect the same files.
Tests are run by Codeship (our CI service) against every commit, so it’s easy to see if a given commit is valid for merging, but there’s no way to know if a commit will (a) merge cleanly, and (b) still pass tests after merging, at least until you merge it. (a) is handled by BitBucket (it shows us if a merge will apply cleanly), but (b) is still a problem.
I was able to leverage the “Test Pipelines” feature of Codeship to make a pipeline that checks if the current branch is
default, and if it is not, then it attempts an automated merge, followed by running all of the tests. It doesn’t send results to Codecov, because the commit that would be covered does not exist yet, but it does report errors if it fails to merge, or fails to run tests correctly.
if [ "$CI_BRANCH" != "default" ] ; then hg merge -r default --tool internal:merge; detox; fi
Notes: we use
mercurial, so the branch name is based on that, as is the merge command. We also use
detox to run tests in parallel.
This catches some big issues we had, where two migrations in one app were created in different branches, and the migrations framework is unable to deal with that. It could also pick up other issues, where a merge results in code that does not work, even though both commits contained fully passing code.
It’s not quite PR testing, because if the target revision changes (ie, a different PR is merged), then it should run the merge-checking pipeline again, but I’m not sure how to trigger this.