On Tue, Mar 26, 2019 at 13:14:48 -0500, Eric Blake wrote:
On 3/26/19 12:41 PM, Peter Krempa wrote:
> On Tue, Mar 26, 2019 at 01:13:44 -0500, Eric Blake wrote:
>> I'm fairly confident that these API are ready to go (that is, I've got
>> qemu code in the wings to implement these API for the qemu driver, as
>> demonstrated at last KVM forum, and it shouldn't be too hard to add
>> support in the test driver to get some 'make check' coverage similar
>> to what I recently added for snapshots). I'm hoping the APIs make it
>> in for 5.2, even if I'm still dealing with review churn on the later
>> parts of my v5 series (there has been a lot of rebasing from earlier
>> review comments, so v5 is currently still the most recent version that
>> I was able to run demos with, although I hope to post the rest of v6
>> soon).
>
> I'm still not persuaded that pushing any API without implementation is a
> good idea. (No, test driver implementation does not count). If the
> consensus of others is that it's good to go then go ahead, but I want to
> voice this concern here.
I do have the qemu implementation for checkpoints fully tested, as well
as working for pull mode backups (the push mode backups is still demo
quality, but as of v5, I was able to get a push mode backup started even
if the events weren't wired up correctly).
That is okay. If there is a working subset with a real driver I don't
object. I just don't see any value in pushing API without a real
implementation from upstream point of view.