On 07/14/2010 09:00 AM, Daniel P. Berrange wrote:
On Tue, Jul 13, 2010 at 10:57:02AM -0400, Darryl L. Pierce wrote:
> On Wed, Jul 07, 2010 at 09:22:55AM -0400, Darryl L. Pierce wrote:
>> Network Device:
>>
>> I've scaled back the APIs on this to have the interface's name and mac
>> address (if one is present). The APIs it would present include methods
>> to configure and destroy the device, start and stop the device, and
>> return the NetCF descriptor.
>
> Some additional notes here. NetCF isn't available on Windows, so I think
> this is the wrong direction to go for networking. Instead, more explicit
> APIs for configuration and management are in order. I'm redoing the APIs
> to have methods to add and remove IPv4 addresses, and to add and remove
> DNS entries.
>
> For future work, we'll have to think about how to define a bridge, how
> to bond devices, etc.
>
> Thoughts?
IMHO this sounds like a bad approach. netcf was written precisely so that
apps don't have to keep reinventing network interface management for
every OS. Regardless of whether you use netcf or not, you have to write
Windows support from scratch. Why add yet more work by reimplementing the
Linux stuff netcf already does too ? Just write the Windows support as part
of netcf, and use existing Linux support.
Yeah, we discussed this off list yesterday and this was a point I brought
up as well. Laine made it clear he didn't want to touch netcf on Windows,
but I agree if we're going to have to implement the functionality we might
as well have it in a common place/library.
Perry