On Tue, May 9, 2017 at 4:11 AM, Oliver Gondža <
ogo...@gmail.com> wrote:
> a PR that purports to fix
> something you have no evidence of being broken referring to a TLA-named
> testsuite you have never heared of and have no easy way to checking for
> yourself
FWIW I have asked people filing such PRs to include some background
information as well as instructions to reproduce the failure locally.
In most (not all) cases this comes down to something like
mvn clean test-compile && mvn -Djenkins.version=2.x.y surefire:test
or even
mvn -Djenkins.version=2.x.y clean test
The last can be kept passing by writing your `Jenkinsfile` as
buildPlugin(jenkinsVersions: [null, '2.x.y'])
for whatever LTS line. There are also some cases where failures are
due to plugin-to-plugin dependency bumps, which `buildPlugin` does not
currently support as a testing option.
As to whether there can be a public Jenkins job running precisely the
same set of tests as CloudBees runs internally, I am doubtful, since
the particular set of core & plugin versions being tested is defined
by the versions promoted in CloudBees products, and having this as a
scheduled job would make little sense since a test run is tied to a
change to that revision matrix.
Now what *would* be extremely useful, if someone cares to take the
time to set it up and maintain it, is a job on
ci.jenkins.io which
periodically (daily?) runs the PCT against some defined but
ever-increasing (transitively closed) set of plugins, using trunk
snapshot versions of core and all included plugins. If failures are
evaluated and assigned promptly, this would alert developers to
compatibility regressions they need to either fix or acknowledge,
before releases are even made (albeit after faulty PRs are merged—a
harder problem to solve I think).