Pseudo-triggers are rather misleading. They always add a post-build step BuildTrigger (which is not a Trigger! rather a Notifier; and really should be a Recorder) to the upstream project. You do not need permission to configure the upstream project to do this. The downstream project is not changed, despite the appearance in the configuration screen.
SECURITY-55 (and SECURITY-109) force anyone configuring a job to have build permission on any specified downstream jobs (whether added by a pseudotrigger or not). JENKINS-16956 reverts that, doing the permission check at runtime instead.
Note that BuildTrigger is even weirder than the above description would imply because its perform method does nothing! It exists solely to be a DependencyDeclarer, thus adding an edge to the project dependency graph. And then in AbstractBuildExecution.cleanUp, BuildTrigger.execute is called, which triggers downstream builds for all dependencies from this upstream project—whether added by BuildTrigger or not—in topological order, regardless of what order those project configuration elements were added in (for example in the publishers section). So for example if you use the Parameterized Build plugin, and add a non-blocking trigger to your publishers section, running it does nothing per se, except insofar as that is marked as a dependency in the graph. As compared to adding a blocking trigger as a build step, which does not affect the dependency graph, and does exactly what it looks like it does. Confused yet?
JENKINS-22397, recently implemented, allows the Build Result trigger to display a node in the dependency graph—but for now it is purely informational: not used during BuildTrigger.execute. Currently that trigger (which is a true trigger) runs on a polling schedule, meaning that there is a delay before the downstream build is scheduled. Probably this should be changed to immediately schedule the downstream build, though the rather complex implementation there is tightly tied to polling.
Perhaps there should be a fresh, simple plugin added implementing a Trigger based on an upstream project’s completion, and a Notifier/Builder which schedules a downstream project, both implementing an informational edge in the dependency graph and performing intuitive access control checks. (If you really want the automated topological sort functionality that Jenkins core offers, and understand the implications, you can continue to use BuildTrigger.)
Pseudo-triggers are rather misleading. They always add a post-build step BuildTrigger (which is not a Trigger! rather a Notifier; and really should be a Recorder) to the upstream project. You do not need permission to configure the upstream project to do this. The downstream project is not changed, despite the appearance in the configuration screen.
SECURITY-55 (and SECURITY-109) force anyone configuring a job to have build permission on any specified downstream jobs (whether added by a pseudotrigger or not).
JENKINS-16956reverts that, doing the permission check at runtime instead.Note that BuildTrigger is even weirder than the above description would imply because its perform method does nothing! It exists solely to be a DependencyDeclarer, thus adding an edge to the project dependency graph. And then in AbstractBuildExecution.cleanUp, BuildTrigger.execute is called, which triggers downstream builds for all dependencies from this upstream project—whether added by BuildTrigger or not—in topological order, regardless of what order those project configuration elements were added in (for example in the publishers section). So for example if you use the Parameterized Build plugin, and add a non-blocking trigger to your publishers section, running it does nothing per se, except insofar as that is marked as a dependency in the graph. As compared to adding a blocking trigger as a build step, which does not affect the dependency graph, and does exactly what it looks like it does. Confused yet?
JENKINS-22397, recently implemented, allows the Build Result trigger to display a node in the dependency graph—but for now it is purely informational: not used during BuildTrigger.execute. Currently that trigger (which is a true trigger) runs on a polling schedule, meaning that there is a delay before the downstream build is scheduled. Probably this should be changed to immediately schedule the downstream build, though the rather complex implementation there is tightly tied to polling.Perhaps there should be a fresh, simple plugin added implementing a Trigger based on an upstream project’s completion, and a Notifier/Builder which schedules a downstream project, both implementing an informational edge in the dependency graph and performing intuitive access control checks. (If you really want the automated topological sort functionality that Jenkins core offers, and understand the implications, you can continue to use BuildTrigger.)