Native Socket Error On OS X (BSDs)

[Kurt Miller]

Hi Xiaobin,

I noticed your comment about removing InterruptibleIO support from the
BSD port. I think that is a good idea. It is one of the areas where I
diverged from the Linux code base a bit to bring it in. Removing it
will make the BSD closer to the Linux code base and further reduce
the bsd-port divergence. I'll look into that after I get synced up
and merge in some other changes I'm working on now.

Regards,
-Kurt

Xiaobin Lu wrote:
> Hi Michael,
> 
> I agree with you that we should understand why the comments are there
> and whether it is correct. I am also wondering whether we should remove
> the interruptible I/O support completely. The interruptible I/O is only
> supported on Solaris and recently we switch the flag
> UseVMInterruptibleIO to false by default for JDK 7 (bug 6554406 &
> 6382902). But it is probably a good idea to remove that support since
> that will get rid of many confusions around.
> 
> There was a discussions around whether we should use "select" over
> "poll" in general and the past experience told us "poll" performs better
> and "select" is problematic and can result in very odd failures if you
> have file handle numbers that exceed the bitmap capacity. Plus, on some
> systems select is emulated via poll.
> 
> Regards,
> -Xiaobin
> 
> On 12/23/08 10:54, Michael Franz wrote:
>> Xiaobin,
>>
>> I would rather determine if the comments are correct or not.
>>     // XXXDARWIN: poll() appears non-interruptable on Leopard:
>> Thread.interrupt() failed to
>>     // cause interrupt. Does poll work at all on Tiger? Needs
>> investigation.
>>
>> I am interested in why Linux and FreeBSD cannot use this code.
>> If we can have the same code for all platforms it will be easier to
>> maintain.
>>
>> Michael
>>
>> On Tue, Dec 23, 2008 at 1:20 PM, Xiaobin Lu <Xiaobin.Lu at sun.com
>> <mailto:Xiaobin.Lu at sun.com>> wrote:
>>
>>     Michael,
>>
>>     Thanks for your investigation on this. So instead of removing the
>>     Apple specific code, can we just detect the "-1" argument and pass
>>     "NULL" to select as the last argument which has equivalent meaning
>>     as "-1" to "poll"? I think that might be a safer fix than just
>>     removing that code.
>>
>>     My 2 cents.
>>
>>     -Xiaobin
>>
>>
>>
> 
> 
> ------------------------------------------------------------------------
> 
>