Possible bug in Resources/Jobs

[basilevs]

RCPTT test creating and deleting an IFile in one thread and reading it concurrently in another has failed once in x86_64 Linux CI targeting Eclipse Platform 2021-12 (it completed about 30 times fine before and I can't reproduce it since). There are many problems with the code under test, but the stack trace of this particular failure has surprised me:

!ENTRY org.eclipse.core.jobs 4 2 2025-11-04 03:56:22.213
!MESSAGE An internal error occurred during: "Delete/create".
!STACK 0
java.lang.IllegalArgumentException: endRule without matching beginRule: P/x
	at org.eclipse.core.runtime.Assert.isLegal(Assert.java:66)
	at org.eclipse.core.internal.jobs.ImplicitJobs.end(ImplicitJobs.java:117)
	at org.eclipse.core.internal.jobs.JobManager.endRule(JobManager.java:726)
	at org.eclipse.core.internal.resources.WorkManager.checkOut(WorkManager.java:170)
	at org.eclipse.core.internal.resources.Workspace.endOperation(Workspace.java:1515)
	at org.eclipse.core.internal.resources.Resource.delete(Resource.java:791)
	at org.eclipse.core.internal.resources.Resource.delete(Resource.java:732)
	at org.eclipse.rcptt.core.persistence.plain.PlainTextPersistenceModelTest.lambda$0(PlainTextPersistenceModelTest.java:91)
	at org.eclipse.core.runtime.jobs.Job$2.run(Job.java:187)
	at org.eclipse.core.internal.jobs.Worker.run(Worker.java:63)

The error seems to be purely internal to the Resources framework. What could my code possibly do to achieve this? The whole product does not call IJobManager.endRule() explicitly anywhere.

The test in question is:

	@Test
	public void returnNullWhenResourceDoesNotExist() throws IOException, CoreException {
		ITestCase file = createTestCase();
		URI uri = toURI(file);
		Scenario scenario = ScenarioFactory.eINSTANCE.createScenario();
		scenario.setVersion(Double.toString(RcpttCore.SCENARIO_VERSION));
		scenario.setId(EcoreUtil.generateUUID());
		scenario.setName("test");
		scenario.setDescription("description1");
		saveLoad(scenario, uri);
		Q7LazyResource resource = new Q7LazyResource(uri);
		
		Job noise = Job.create("Delete/create", (ICoreRunnable) monitor -> {
			while (!monitor.isCanceled()) {
				file.getResource().delete(true, monitor);
				saveLoad(scenario, uri);
			}
		});
		noise.setPriority(Job.INTERACTIVE);
		noise.schedule(0);
		try {
			for (long stop = currentTimeMillis() + 1000; currentTimeMillis() < stop;) {
				IPersistenceModel model = PersistenceManager.getInstance().getModel(resource);
				if (model != null) {
					model.updateMetadata(); // should not throw
				}
			}
		} finally {
			noise.cancel();
		}
	}

Which translates to Jobs and Resources as:

	@Test
	public void ensureResourcesSurviveConcurrency() throws IOException, CoreException {
		IProject project = WORKSPACE.getRoot().getProject("project");
		project.create(null);
		project.open(null);
		IFile file = project.getFile("file.
		file.create(new ByteArrayInputStream("content".getBytes()), 0, null);
		Job noise = Job.create("Delete/create", (ICoreRunnable) monitor -> {
			int i = 0;
			while (!monitor.isCanceled()) {
				file.delete(true, monitor);
				file.create(new ByteArrayInputStream(("content " + i).getBytes()), IResource.FORCE, null);
			}
		});
		
		noise.setPriority(Job.INTERACTIVE);
		noise.schedule();
		long iteration = 0;
		try {
			for (long stop = currentTimeMillis() + 10_000; currentTimeMillis() < stop; iteration++) {
				try (InputStream stream = file.getContents(true)) {
						stream.readAllBytes();
				} catch (CoreException e) {
					// Ignore file not found exception
					int code = e.getStatus().getCode();
					if (code == IResourceStatus.RESOURCE_NOT_LOCAL || code == IResourceStatus.RESOURCE_NOT_FOUND || code == IResourceStatus.FAILED_READ_LOCAL) {
						continue; // File does not exist errors are expected
					}
					throw new AssertionError("Error code: " + code, e);
				}
			}
		} catch (Throwable e) {
			throw new AssertionError("Failed on iteration " + iteration, e);
		} finally {
			noise.cancel();
		}
	}

There are resource listeners doing some resources reading too (without endRule()).

[laeubi]

Copilot thinks it is a bug

• Fix race condition in WorkManager causing "endRule without matching beginRule" errors laeubi/eclipse.platform#25

if you can confirm I can open a pullrequest, for this fetch the PR, build the jar locally and replace it in your installation (make sure to use a different version for the bundle and add a printout to see it is effective).

[basilevs]

I can't reproduce the bug, so would not be able to confirm. I've only noticed it once so far.

[basilevs]

I'm going to add test matrix to RCPTT to ensure that latest platform gets coverage from RCPTT's tests. Then if the failure happens again, at least the results would not be obsolete.