RFR: 4833719: (bf) Views of MappedByteBuffers are not MappedByteBuffers, and cannot be forced [v4]

[Brian Burkhalter]

On Tue, 16 Mar 2021 14:39:19 GMT, Alan Bateman <alanb at openjdk.org> wrote:

>> @AlanBateman Hmm, I think that's a different issue to my comment.
>> 
>> I asked my question above because the correction Brian made when documenting slice(int index, int length) employed a half-closed interval with an exclusive right bound for the interval
>> 
>>     [index,index+length)
>> 
>> and I merely felt that the same should apply here in the documentation of slice()
>> 
>>   [position(),limit())
>> 
>> I'm not sure I am in a position to provide the right answer to your question but I'll note that in the same pedantic vein the intervals you mention should be similarly semi-closed
>> 
>>     [0, capacity)
>> 
>> and
>> 
>>     [0, limit)
>> 
>> The byte at index limit/capacity is not necessarily flushed (modulo page roundings).
>> 
>> As to what might be a correct answer:
>> 
>>> The no-arg force method used to write back the entire buffer content [0, capacity]. It has been changed to use the limit, meaning [0, limit]. I don't recall this being an intentional change.
>> 
>> I believe this was done intentionally when I introduced the force(int,int) variant and re-implemented the noargs variant to indirect to it. I cannot recall why it changed from capacity to limit though.
>
> @adinn My comment wasn't to you, it's a concern that the no-arg force method has changed and the new behavior is proposed to be specified by this PR. I need to re-read the mails as it may be that this will need to be restored to use the capacity.

The right hand limit should in all cases be open so the current no-arg doc is wrong.

As for `force()` changing from `[0,capacity())` to `[0,limit())`, I don't know exactly why that happened or when but it definitely changed since Java 8.

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

PR: https://git.openjdk.java.net/jdk/pull/2902