On Fri, Oct 12, 2018 at 00:10:02 -0500, Eric Blake wrote:
The following is the latest version of my API proposal for incremental backups.
I have even more work-in-progress patches on top of these: https://repo.or.cz/libvirt/ericb.git which I am slowly improving to be more in line with my thread on the overview of the API usage: https://www.redhat.com/archives/libvir-list/2018-October/msg00217.html
But I am fairly satisfied that the API as presented is sufficient for everything I have still been implementing in the qemu driver, and that even when qemu is slightly tweaked (such as dropping the x- prefix on various commands, or maybe adding a new command to make it easier to compute the estimated size of the union of several bitmaps), those changes will be limited to the src/qemu directory rather than affecting the API.
Since I will be demonstrating the use of this API at the KVM Forum, I would really like a decision on whether we can commit the API into libvirt now, even if we have to wait for the qemu implementation
No we can't. We did it once in the past and although I don't remember the exact details I remember it turned out to be a bad idea. Also pushing an API with no implementation doesn't really make sense since the API is unusable until it is implemented in some hypervisor driver.
of the API until qemu stabilizes its interfaces (also, having the libvirt API in place gives qemu an incentive to drop the x- prefix sooner rather than later).
The presence of public API in libvirt gives no clue to QEMU engineers that their API provides everything needed and x- prefix should be dropped. Only working patches implementing the libvirt APIs for QEMU can provide such feedback to QEMU developers. Jirka