[kaffe] NetworkInterfaceTest failed

[Dalibor Topic]

Dalibor Topic wrote:
> Guilhem Lavaux wrote:
> 
>> Timothy Stack wrote:
>>
>>>>
>>>>
>>>> Now NetworkInterface uses a vector for keeping network interfaces.
>>>> So both [I, {A1}] and [I, {A2}] are stored.  This is why
>>>> both [I, {A1}] and [I, {A2}] are returned now.
>>>>
>>>> I think we should somehow create [I, {A1, A2}] from
>>>> ([I, {A1}], [I, {A2}]).
>>>
>>>
>>>
>>>
>>> Sorry, I just can't help myself...  The original JanosVM 
>>> implementation of this did "the right thing", i think.  It used a 
>>> hash table to store each network interface and then added the ip 
>>> addresses to the vector in the object.  Lets not overwrite 
>>> functioning code with broken stuff, okie... *grumble*
>>>
>>> tim
>>>
>>
>> Keep calm. ;) First let's rewrite something solid using HashTable. 
>> Then we'll try to push it forward on classpath, hoping that that time 
>> they will accept the modification. :)
> 
> 
> I've talked about it with Michael Koch on IRC on #kaffe, who's familiar 
> with the gcj side of the networking implementation. It seems that gcj 
> does not support IPv6 at all, and noone else using Classpath apparently 
> does, so the implementation in Classpath has not really been exposed to 
> such an interesting usage case (one NetworkInterface name, two 
> InetAddresses), thus it actually is broken in that respect.
> 
> I'll write up some code that 'condenses' the NativeInterfaces that are 
> received from the native method, and I hope that will fix the breakage.

Done & checked in. Ito, could you give it a try? If it works for you, 
then I'll send it upstream to GNU Classpath.

cheers,
dalibor topic