RFR: 8305744: (ch) InputStream returned by Channels.newInputStream should have fast path for ReadbleByteChannel/WritableByteChannel [v3]

[Markus KARG]

On Tue, 20 Jun 2023 22:15:13 GMT, Brian Burkhalter <bpb at openjdk.org> wrote:

>>> ```java
>>> @Benchmark
>>> public void ChannelInputStream_transferTo() throws Throwable {
>>>     var is = Channels.newInputStream(Channels.newChannel(new ByteArrayInputStream(new byte[1024 * 1024 * 1024])));
>>>     var os = Channels.newOutputStream(Channels.newChannel(new ByteArrayOutputStream(1024 * 1024 * 1024)));
>>>     is.transferTo(os);
>>> }
>>> ```
>> 
>> The above benchmark method includes the construction of 6 objects in the measurement. I think it would be more representative of the effect of the change to do something like this:
>> 
>> 
>>     private static final int SIZE = 1024*1024*1024; 
>>  
>>     ByteArrayInputStream in = new ByteArrayInputStream(new byte[SIZE]);
>>     ByteArrayOutputStream out = new ByteArrayOutputStream(SIZE);
>>  
>>     ReadableByteChannel rbc = Channels.newChannel(in);
>>     WritableByteChannel wbc = Channels.newChannel(out);
>> 
>>     InputStream is = Channels.newInputStream(rbc);
>>     OutputStream os = Channels.newOutputStream(wbc);
>> 
>>     @Setup
>>     public void setup() {
>>         in.mark(SIZE + 1);
>>     }
>> 
>>     @Benchmark
>>     public void ChannelInputStream_transferTo() throws Throwable {
>>         is.transferTo(os);
>>         in.reset();
>>         out.reset();
>>     }
>
>> @Setup
>>     public void setup() {
>>         in.mark(SIZE + 1);
>>     }
> 
> In my benchmark example the above code is unneeded.

@bplb I would be glad if we could continue with the review of this PR.

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

PR Comment: https://git.openjdk.org/jdk/pull/13387#issuecomment-1662494403