[PATCH v1 05/26] qemu_domain_address: Reformat qemuDomainAssignS390Addresses()

[Cornelia Huck]

On Mon, 30 Nov 2020 11:18:20 +0100
Michal Privoznik <mprivozn at redhat.com> wrote:

> On 11/30/20 10:38 AM, Thomas Huth wrote:
> > On 27/11/2020 16.02, Michal Privoznik wrote:  
> >> Signed-off-by: Michal Privoznik <mprivozn at redhat.com>
> >> ---
> >>   src/qemu/qemu_domain_address.c | 10 ++++------
> >>   1 file changed, 4 insertions(+), 6 deletions(-)

> >>       } else if (virQEMUCapsGet(qemuCaps, QEMU_CAPS_VIRTIO_S390)) {  
> > 
> > Not related to your patch, but an idea for a future clean-up: That
> > QEMU_CAPS_VIRTIO_S390 seems to belong to the ancient "s390-virtio" (without
> > ccw) machine that has been removed in QEMU v2.6 already:
> > 
> > https://git.qemu.org/?p=qemu.git;a=commitdiff;h=7b3fdbd9a82
> > https://git.qemu.org/?p=qemu.git;a=commitdiff;h=3538fb6f89d
> > 
> > IIRC, that machine was already considered as deprecated since a couple of
> > earlier QEMU releases, so I really doubt that anybody is still using that in
> > production today.
> > 
> > Thus I think that all code related to QEMU_CAPS_VIRTIO_S390 could likely be
> > removed from libvirt nowadays.  
> 
> That is even better idea. But currently libvirt supports QEMU-1.5.0 and 
> newer. So I think we shouldn't remove that until the minimum version is 
> bumped even though we think feature has no users.
> 
> https://gitlab.com/libvirt/libvirt/-/commit/c1bc9c662b4
> 
> Although, it might be about time to look again what is the oldest QEMU 
> we need to support.

Would be great if you could bump it enough to get rid of the old
virtio-s390 transport :)

FWIW, virtio-ccw was introduced in QEMU 1.4, and became the default
with QEMU 2.4, although it had supplanted virtio-s390 well before that.
What are the criteria for possibly removing support for a feature in
libvirt: that nobody would use it in practice, or that nobody would be
able to use it?