RFR 8058824: Drop TwoStacks socket implementation in jdk9 [win] - Part I

Michael McMahon michael.x.mcmahon at oracle.com
Tue Sep 23 11:06:04 UTC 2014


Ivan,

Okay. I see the "part 1" in the subject line now, but maybe the bug 
should be updated
to define the scope of the change and we should file a separate bug 
report then.
I'm okay with leaving the class with the current name. What about the 
datagram socket code?
It might make sense to include the equivalent changes there also. There 
might need to
be some code in net_util_md.c that needs to be removed also 
(NET_Timeout2() function?)

Thanks
Michael

On 23/09/14 11:49, Ivan Gerasimov wrote:
> Thanks Michael!
>
> On 23.09.2014 13:41, Michael McMahon wrote:
>> Hi Ivan,
>>
>> Did you look at the possibility of removing the TwoStacks class 
>> altogether?
>> For Solaris/Linux etc. ipv4 only and ipv6/v4 are all handled in the 
>> same impl class
>> with just a switch at socket creation time, selecting AF_INET or 
>> AF_INET6.
>>
> Yes, trying to merge DualStack with IPv4-only (the remaining part of 
> TwoStacks) was planned as the second step.
> It's not that straight-forward, so better be done separately from this 
> change.
>
>> If there is a good reason to keep the implementations separate then I 
>> think
>> the class should be renamed to reflect its new purpose.
>>
> I left it with the old name to make the review simpler.
> No problem to rename the files, of course.
> I only thought that if we are planing to combine both implementations, 
> renaming will not matter that much and only make the things more 
> complicated.
>
> Sincerely yours,
> Ivan
>



More information about the net-dev mailing list