On 09/01/2015 02:11 PM, Laine Stump wrote:
On 08/31/2015 04:06 PM, Jonathan Toppins wrote:
> On 08/31/2015 03:25 PM, Guido Günther wrote:
>> Hi,
>> On Sat, Aug 29, 2015 at 04:19:10PM -0400, Jonathan Toppins wrote:
>>> Adds a new interface type using UDP sockets, this seems only applicable
>>> to QEMU but have edited tree-wide to support the new interface type.
>>>
>>> The interface type required the addition of a "localaddr" (local
>>> address), this then maps into the following xml and qemu call.
>>>
>>> <interface type='udp'>
>>
>>
>> Sine we do have
>>
>> <interface type='mcast'>
>>
>> already wouldn't it be better to have something like
>>
>> <interface type='ucast' protocol='udp'>
>>
>
> This possibly could be better, my concern would be now tcp is
> configured differently than udp, no? Or are you saying something like:
>
> <interface type='ucast' protocol='udp|tcp'>
I think the case of a tcp connection is already handled by <interface
type='client'> and <interface type='server'> together, so that
doesn't
seem likely to happen. I suppose it's possible someone would come up
with an sctp-based transport in the future though. I'm undecided about
this.
I am willing to work with Guido if you have some time to put into it.
Otherwise the current implementation <interface type="foo"> models are
pretty much 1:1 representations of what is in the C code and this looks
pretty extensible to me. Have not seen any (or very few) copy-and-paste
sections of code so it seems to be holding up currently. I will not try
and speak for supporting the xml configuration long term, would
appreciate some other perspectives. Lacking this I leave it up to the
maintainer(s) to determine if this is worthy to go in. I see v1.2.19-rc2
has been put out there so a possible compromise (policy even?) is for
this to go into v1.2.20-devel so people can play around with it. I
assume implementations can be changed/reverted between releases v1.2.19
to v1.2.20. Completely understand the concern about supporting an
interface for ever :)
Regards,
-Jon