Igor Mammedov <imamm...@redhat.com> writes: ... > > i was talking about user interface here, i.e.: > (QMP) create_machine(hyperv=on) > (QMP) device_add(cpu, hv_foo=x) > vs: > (QMP) device_add(cpu, hyperv_defaults=on,=onhv_foo=x) > > i.e. in the later case cpu specific options are consolidate within device > stanza > and mgmt doesn't need to be aware and split cpu config in to steps. >
FWIW, 'hv_foo=x' only makes sense if 'x' == 'off' as currently we enable all existing enlightenments. Also, this requires upper layer tools to know something about 'hv_foo' (to be able to disable it) and AFAICT layers above libvirt don't actually want to know such low-level details. Don't get me wrong, I'm not against 'hv_defaults=on', just trying to give the perspective so we won't need to change the interface again anytime soon. -- Vitaly
