RFR: 8320971: Use BufferedInputStream.buf directly when param of implTransferTo() is trusted [v15]

[Vladimir Sitnikov]

On Mon, 4 Dec 2023 14:05:14 GMT, Vladimir Sitnikov <vsitnikov at openjdk.org> wrote:

>>> > it doesn't solve the second part of the concern which is read-only view on the internal buffer.
>>> 
>>> I think `ByteBuffer.asReadOnlyBuffer()` should resolve the issues regarding naked byte access. At least `ByteBuffer#array()` returns `null` for read-only buffers, and `ByteBuffer#hasArray()` returns `false`. Am I missing something?
>> 
>> The backing array is not frozen so a read-only view doesn't address the second part of the concern where the target keeps a reference.
>
>>The backing array is not frozen so a read-only view doesn't address the second part of the concern where the target keeps a reference.
> 
> The users outside of OpenJDK would not have a reference to `byte[]`.
> Do you mean a reference to `ByteBuffer`?
> What if there was a `ByteBuffer#forbidReads()` method that would permanently forbid subsequent reads from the `ByteBuffer`?

I've created a draft change for `OutputStream#write(ByteBuffer)`, and I have benchmarked several cases including `ByteArrayInputStream.transferTo(BufferedOutputStream(ByteArrayOutputStream))`, `ByteArrayInputStream.transferTo(DataOutputStream(FileOutputStream))`.

Benchmarks show there's improvement in both allocation rate and latency.

Read-only `ByteBuffers` can address poisoning concerns as non-jdk code can't peek into read-only arrays.

The `write(ByteBuffer)` approach supports cases like `CheckedOutputStream`, `DataOutputStream` which would be hard or impossible to optimize when passing naked byte arrays.

Here are the results: https://gist.github.com/vlsi/7f4411515a4f2dbb0925fffde92ccb1d
Here is the diff: https://github.com/vlsi/jdk/compare/ce108446ca1fe604ecc24bbefb0bf1c6318271c7...write_bytebuffer

@mkarg , @bplb , @AlanBateman , could you please review `OutputStream.write(ByteBuffer)` approach?

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/16879#discussion_r1428725435