Failure from BuildListener are ignored

[ljacomet]

When a BuildListener is registered, any exception thrown during its execution should fail the build.

With the following settings.gradle:

def adapter = new BuildAdapter() {
    void projectsEvaluated(Gradle gradle) {
        println 'About to break the build!'
        throw new GradleException('BOOOM')
    }
}

gradle.addListener(adapter)

and an empty build.gradle

Expected Behavior

Running ./gradlew tasks should fail the build as the listener throws an exception.
That is the case with Gradle 4.9

Current Behavior

Starting with Gradle 4.10 and the latest 5.0 nightly, the build is successful.

Steps to Reproduce (for bugs)

• Create a settings.gradle as indicated above
• Create an empty build.gradle
• Run with Gradle > 4.10 and see build pass
• Run with Gradle <= 4.9 and see build fail

Your Environment

• Build scan URL: https://e.grdev.net/s/5vhex5mtfuyba

[ljacomet]

From initial investigations, it looks like between this code which sets the failure on the context and this code which creates the context and passes it down for execution, nowhere is the context queried for a failure and that failure thrown out.

[mkobit]

Maybe similar to #6900 ?

[ljacomet]

I should have searched better, thanks for the link!
But the difference between Kotlin and Groovy DSL might indicate a different problem.
So not marking this one a duplicate just now.

[ljacomet]

Playing around with the behavior reported in #6900, I tested using Gradle.projectsEvaluated and both the Closure and Action based version do fail the build when used.

[ldaley]

@ljacomet I think the solution should be to change

gradle/subprojects/core/src/main/java/org/gradle/configuration/internal/DefaultListenerBuildOperationDecorator.java

Line 243 in 2c68792

} catch (Exception e) {

to just throw there (by using JavaMethod to do the invoke).

And… add some test coverage :S

[oehme]

The decoration did not cause this failure - It doesn't matter if you throw or call context.setFailure, the exception will be propagated.

The reason why buildFinished can no longer fail the build are the changes in feb84b4. This was not caught, because the corresponding test case has been @Ignored for a long time.

@adammurdoch would you mind taking a look at this? The behavior is pretty confusing now, as you will get both a Build Successful and Build Failed message in the console output and the latter lacks the correct location information. The overall return code of Gradle is successful.

Looking at the code, the behavior was odd before too. If a listener failed, it was actually called again with that failure.

[slaverda]

Hi All,

Do you have any updates on that issue ? It's blocking our migration to the latest gradle.

[adammurdoch]

Fixed via #7510. The fix will be in Gradle 5.0.

The PR doesn't fix the issue of missing location information and misleading BUILD SUCCESSFUL message for buildFinished failures. That is a separate issue that I'll fix in master for 5.1, as it's been around for a very long time.