Buffers.clear(...) only zeros the buffer's own window of the backing array

[kdelay]

What

Buffers.clear(XBuffer) promises to clear this buffer. For array-backed buffers it does more than that: every overload calls

Arrays.fill(buffer.array(), 0);

array() returns the whole backing array, so arrayOffset() and capacity() are ignored. A buffer returned by slice() shares its parent's array at a non-zero offset, so clearing a slice also zeros bytes that belong to the parent and are outside the slice.

The direct branch of the same method is correct, because it writes zeros through the buffer and therefore respects the window. So the two branches of one method disagree.

Reproduction

JDK 17, master at eb60bd5, 10-element heap array filled with 7, slice over [4, 7):

byte[] backing = new byte[10];
Arrays.fill(backing, (byte) 7);
ByteBuffer parent = ByteBuffer.wrap(backing);
parent.position(4);
parent.limit(7);
ByteBuffer slice = parent.slice();   // capacity 3, arrayOffset 4

Buffers.clear(slice);

path	result
array-backed slice	[0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
same test with allocateDirect	[7, 7, 7, 7, 0, 0, 0, 7, 7, 7]

The direct result is what the Javadoc describes.

Change

clearBuffer() has already set the position to zero and the limit to the capacity, so the region that belongs to this buffer is [arrayOffset(), arrayOffset() + capacity()). This switches the seven array-backed branches to the ranged Arrays.fill() overload. Buffers with arrayOffset() == 0 are unaffected, which covers everything allocate() and wrap() produce, and the direct branches are untouched.

Buffers is new in 2.23.0 and is not released yet, and the only in-tree callers go through clearDirect(...), so this does not change behavior anyone can be depending on today.

Tests

Four regression tests in BuffersTest (byte, char, int, short, covering each cast variant of the fill). Verified they are load-bearing:

• with the change: Tests run: 61, Failures: 0, Errors: 0, Skipped: 0
• with Buffers.java reverted: Tests run: 61, Failures: 4

Full default goal (mvn, so rat, checkstyle, japicmp, spotbugs, pmd and javadoc included): BUILD SUCCESS, Tests run: 6414, Failures: 0, Errors: 0, Skipped: 36.

---

• Read the contribution guidelines for this project.
• Read the ASF Generative Tooling Guidance if you use Artificial Intelligence (AI).
• I used AI to create any part of, or all of, this pull request. Which AI tool was used to create this pull request, and to what extent did it contribute?
  • Tool: Claude Code (Anthropic), running as an autonomous contribution workflow.
  • Extent: all of it. The defect was found by diffing the @since 2.23.0 Javadoc contracts against the implementations, and the probe, the patch, the tests and this description were written by the tool.
  • Verification: every number quoted above comes from a run on this machine, not from the model. The probe was executed before and after the patch, and the reverted-source run is how the four tests were confirmed to fail without the change.
  • Human oversight: none before submission. Please review it as unreviewed work. I will answer any question about the change on this thread.
• Run a successful build using the default Maven goal with mvn; that's mvn on the command line by itself.
• Write unit tests that match behavioral changes, where the tests fail if the changes to the runtime are not applied.
• Write a pull request description that is detailed enough to understand what the pull request does, how, and why.
• Each commit in the pull request should have a meaningful subject line and body.

[garydgregory]

I converted this PR to draft.

Human oversight: none before submission. Please review it as unreviewed work. I will answer any question about the change on this thread.

This is not acceptable. Do your own homework before dumping it on someone else.