@BeforeStep/@AfterStep methods are ignored by ChunkOrientedStepBuilder#listener(Object)

[acktsap]

I searched existing issues and found related listener-registration issues such as #5092, #5094, and #5347, but I could not find a duplicate for this specific case.

Bug description

ChunkOrientedStepBuilder#listener(Object) ignores annotation-based listeners that only declare @BeforeStep or @AfterStep methods.

The same object is registered if it also has another chunk/item listener annotation, such as @BeforeRead.

Environment

Spring Batch: 6.0.4
Java: 21

Steps to reproduce

Register a POJO with only a step-level listener annotation on a chunk-oriented step:

AtomicInteger called = new AtomicInteger();

Object listener = new Object() {

    @BeforeStep
    public void beforeStep(StepExecution stepExecution) {
        called.incrementAndGet();
    }

};

ChunkOrientedStep<String, String> step = new StepBuilder("step", new ResourcelessJobRepository())
        .<String, String>chunk(1)
        .reader(new ListItemReader<>(List.of("item")))
        .writer(chunk -> {
        })
        .listener(listener)
        .build();

step.execute(new StepExecution("step", new JobExecution(1L, new JobInstance(1L, "job"), new JobParameters())));

assertThat(called).hasValue(1); // currently 0

Expected behavior

@BeforeStep and @AfterStep methods should be detected by ChunkOrientedStepBuilder#listener(Object).

Minimal Complete Reproducible example

The snippet above reproduces the issue.

[seonwooj0810]

Would the team be open to a PR for this?

The root cause looks self-contained: ChunkOrientedStepBuilder#listener(Object) scans for the chunk/item annotations (@BeforeChunk, @BeforeRead, @BeforeWrite, …) but not for the step-level @BeforeStep/@AfterStep, so a POJO declaring only those produces an empty method set and is never registered. StepBuilderHelper#listener(Object) already handles them (it adds BeforeStep.class/AfterStep.class to its scan), so this seems to be an inconsistency introduced by the new builder rather than an intentional change.

If a fix is welcome, I'd add the two missing annotations to the scan and cover it with a regression test based on the reporter's reproducer.