Build MethodSource with the correct spec class name.

[milo2048]

The combination of Spock, Junit5, and the maven Surefire plugin prevents tests from running if they're inherited from a parent class.

Consider this scenario:

class MyParentSpec extends Specification {
    def 'some test'() {
        when: 'A thing happens'
        def foo = 'someValue'
        then: 'It works'
        assert foo == 'someValue'
   }
}

class MyChildSpec extends MyParentSpec() {
}

Run command line with:
mvn -Dtest=MyChildSpec

No tests will run. Ultimately this is because Spock reports a test plan to Junit5 wherein the test 'some test' belongs to MyParentSpec instead of MyChildSpec, and the Surefire plugin then filters out the test.

The same kind of inheritance scenario with a pure Junit5 test works as expected.

From the JUnit5 docs for MethodSource:

/**
	 * Create a new {@code MethodSource} using the supplied
	 * {@link Class class} and {@link Method method}.
	 *
	 * <p>This method should be used in favor of {@link #from(Method)} if the
	 * test method is inherited from a superclass or present as an interface
	 * {@code default} method.
	 *
	 * @param testClass the Java class; must not be {@code null}
	 * @param testMethod the Java method; must not be {@code null}
	 * @since 1.3
	 */
	@API(status = STABLE, since = "1.3")
	public static MethodSource from(Class<?> testClass, Method testMethod) {
		return new MethodSource(testClass, testMethod);
	}

Therefore I believe that Spock should be reporting 'some test' as belonging to MyChildSpec.

With the supplied patch, the tests run through maven/surefire.

[leonard84]

Can you add tests for this, maybe by using the EngineTestKit ?

[milo2048]

That’s tough — the problem is in the interaction between Spock, Junit, and Surefire. The class that doesn’t handle Spock’s metadata is private in the Surefire package so I can’t use it in a test. I pushed a commit with the test, but I had to copy+paste the Surefire plugin’s filter into my test. The test does fail when run against the code in master, and pass in my branch.

…

On Jun 10, 2021, at 3:21 PM, Leonard Brünings ***@***.***> wrote: Can you add tests for this, maybe by using the EngineTestKit ? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1345 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AH3HHZHTXAMS6ZZZ4NTGPPLTSEGD5ANCNFSM46O5NJAA>.

[codecov]

Codecov Report

Merging #1345 (7cc0a98) into master (567b8b0) will increase coverage by 79.20%.
The diff coverage is 100.00%.

@@              Coverage Diff              @@
##             master    #1345       +/-   ##
=============================================
+ Coverage          0   79.20%   +79.20%     
- Complexity        0     3940     +3940     
=============================================
  Files             0      408      +408     
  Lines             0    12253    +12253     
  Branches          0     1616     +1616     
=============================================
+ Hits              0     9705     +9705     
- Misses            0     1967     +1967     
- Partials          0      581      +581     

Impacted Files	Coverage Δ
...ain/java/org/spockframework/runtime/SpockNode.java	76.59% <100.00%> (ø)
.../org/spockframework/runtime/SpockTimeoutError.java	100.00% <0.00%> (ø)
.../org/spockframework/mock/EmptyOrDummyResponse.java	85.00% <0.00%> (ø)
...main/groovy/spock/util/EmbeddedSpecCompiler.groovy	94.59% <0.00%> (ø)
.../org/spockframework/builder/SetterSlotFactory.java	66.66% <0.00%> (ø)
...va/org/spockframework/runtime/model/ErrorInfo.java	100.00% <0.00%> (ø)
...rk/mock/constraint/WildcardArgumentConstraint.java	66.66% <0.00%> (ø)
...ock/util/SourceToAstNodeAndSourceTranspiler.groovy	76.87% <0.00%> (ø)
...java/org/spockframework/runtime/ValueRecorder.java	73.07% <0.00%> (ø)
...ava/org/spockframework/runtime/model/NodeInfo.java	100.00% <0.00%> (ø)
... and 399 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 567b8b0...7cc0a98. Read the comment docs.

[Vampire]

This is a bug in the surefire plugin that has to be fixed there I think.
Gradle had the same issue and fixed it.
Quoting @marcphilipp who is both a JUnit and Gradle developer from #1235:

The MethodSource is supposed to contain the declaring class, hence "source", and was introduced to allow IDE navigation. The filtering code could check for the parent descriptor's ClassSource in this case instead of using the MethodSource's class name.

[Vampire]

Here the PR that fixed it in Gradle: gradle/gradle#15943

[leonard84]

Good point @Vampire, I completely forgot about that issue with gradle.

[leonard84]

I talked with @marcphilipp again, and he corrected is statement from the other PR and mentioned this commit junit-team/junit5@8856b48

[leonard84]

I've simplified the PR and removed the maven dependency, while preserving the effect of the maven test filter in the test.

[leonard84]

Thanks @milo2048