RFR 8245194: Unix domain socket channel implementation

[Michael McMahon]

On 12/09/2020 17:44, Alan Bateman wrote:
> On 11/09/2020 16:55, Michael McMahon wrote:
>>
>> On 11/09/2020 15:46, Alan Bateman wrote:
>>> On 26/08/2020 15:24, Michael McMahon wrote:
>>>> :
>>>>
>>>> There is a fair bit of other refactoring and simplification done 
>>>> also. Next version is at:
>>>>
>>>> http://cr.openjdk.java.net/~michaelm/8245194/impl.webrev/webrev.9/
>>> I realise this has moved to a PR but I'm still working through some 
>>> of the areas that are touched in this patch.
>>>
>>> Switching the Pipe implementation to use a Unix domain socket on 
>>> Windows is good but it looks like it will try to create a Unix 
>>> domain socket every time, even on editions of Windows that don't 
>>> support it. Can this be changed to try once and fallback to the 
>>> existing loopback connection approach thereafter? Minor nits here 
>>> but "getServer" returns a listener socket so needs a better name. 
>>> Also in the SinkChannelImpl then it would be better to keep the 
>>> layout consistent and put setNoDelay at the end rather than before 
>>> the constructor.
>>>
>> Good points. Would getListener be okay instead of getServer?
> Or createListener as that better describes what it does. One thing I'm 
> wondering about is adding a new constructor to the Windows PipeImpl 
> with a flag to indicate if it should not buffer, meaning Nagle is 
> disabled when it does a loopback connection. That would avoid the need 
> for setNoDelay in the Windows Selector code, a welcome cleanup that 
> we'll need to do anyway once we have another Selector implementation 
> for this platform.
>
Okay, that is fine. I just pushed those changes to the branch. They can 
be seen in webrev.03 in the PR (when it appears).

Thanks,

Michael.

> -Alan