Note that this also affects repositories that need to be recovered (using 'hg recover'). Currently Jenkins just builds the last change that was pulled before the repository got in a bad state. This way it is very hard to see that this is the issue. Checking the exit code and stopping would make it easy to replay by hand at the right point in the build.

In the current state, this bug totally undermine the whole notion of CI and automatic testing. Because teams, which rely on Jenkins checking out and verifying the latest changes, will silently get some arbitrary older version verified. Which may lead to wrong decisions and wrong versions being deployed to the production. With potentially huge business impact.

Code changed in jenkins
User: Jesse Glick
Path:
src/main/java/hudson/plugins/mercurial/MercurialSCM.java
src/test/java/hudson/plugins/mercurial/SCMTestBase.java
http://jenkins-ci.org/commit/mercurial-plugin/0b005bd18419f81ecb2d98827054273d9d70fbdc
Log:
Merge pull request #56 from ben--/master

[FIXED JENKINS-15806] - Fail build on failed pull

Compare: https://github.com/jenkinsci/mercurial-plugin/compare/c064ebaab7fd...0b005bd18419

Since upgrading to 1.50, when my jobs get triggered for any reason other than a commit to the repository (e.g. an upstream trigger or a manual build) they get failed by the Mercurial plugin because the hg pull found no changes:

Started by user Andrew Johnson
[EnvInject] - Loading node environment variables.
Building on master in workspace /var/lib/jenkins/jobs/epics-pvaPy/workspace
[workspace] $ hg showconfig paths.default
[workspace] $ hg pull --rev default
pulling from http://hg.code.sf.net/p/epics-pvdata/pvaPy
no changes found
ERROR: Failed to pull
ERROR: Failed to pull
[WARNINGS] Skipping publisher since build result is FAILURE
Sending e-mails to: anj@aps.anl.gov
Notifying upstream projects of job completion
Finished: FAILURE

To get them to build in this state I have to wipe out the work-space so they get cloned again. This is a major problem which I'm pretty sure this change was responsible for, so I will be reverting to 1.49.

@anjohnson please file any follow-up issues separately and link to the original bug fix. Generally reopening an issue should be reserved for the case that the fix somehow missed correcting the original problem at all.

In this case the problem is likely that you have an older Mercurial installation which improperly reports a nonzero exit status when pull has nothing to pull. Upgrade it.