6916202: (so) improve scatter/gather implementation

[Chris Hegarty]

Alan,

To clarify ( sent in a private mail ), the bug number is 6971825 and the 
webrev is:
   http://cr.openjdk.java.net/~alanb/6971825/webrev

Comments:

Util.getTemporaryDirectBuffer. If reading/writing in non 
scattering/gathering it looks like a new buffer will be created every 
time. Should L64 be i <= count?

Indentation for "throw new IndexOutOfBoundsException();" in 
DatagramChannelImpl and FileChannelImpl seems to be only three spaces.

The temporary buffer cache is no longer SoftRefs. Do you see any 
potential side effect of this? Is this deliberate?

IOUtil.write/read. Releasing the shadow buffer and clearing the vec refs 
could just be done in the finally block with need for completed. But 
then your committing to a second loop, where as with what you have the 
second loop is only necessary if there is a failure, right?

It looks like now you can only read/write with a limit of 8 buffers, 
where as before it just spilled over to creating short lived direct 
buffers. I think this should be sufficient, just an observation that we 
don't spill anymore.

-Chris.

Alan Bateman wrote:
> Chris Hegarty wrote:
>> I think you used the wrong bug number, 6971825 right?
>>
>> -Chris.
> 6916202 was one I was looked at with Vinnie last night :-)
> 
> Here's the webrev:
>  http://cr.openjdk.java.net/~alanb/6971825/webrev/
> 
> Just so you know, this one is for 6u22m but I want to get it into jdk7 
> and let it bake for a bit first.  I've asked Sherman to review it too 
> (the more eyes the better as the changes are tricky).