On 05/16/2018 06:30 PM, John Ferlan wrote:
>
>
> On 05/16/2018 11:09 AM, Pavel Hrdina wrote:
> > On Wed, May 16, 2018 at 10:52:56AM -0400, John Ferlan wrote:
> > > Using a QEMU 2.12 tagged tree build and full build, used:
> > >
> > > tests/qemucapsprobe /home/qemu/x86_64-softmmu/qemu-system-x86_64 > \
> > > tests/qemucapabilitiesdata/caps_2.12.0.x86_64.replies
> > >
> > > VIR_TEST_REGENERATE_OUTPUT=1 tests/qemucapabilitiestest
> > > VIR_TEST_REGENERATE_OUTPUT=1 tests/domaincapstest
> > >
> > > Signed-off-by: John Ferlan <jferlan(a)redhat.com>
> > > ---
> > > Reposting :
> > >
> > >
https://www.redhat.com/archives/libvir-list/2018-May/msg00738.html
> > >
> > > with recent updates through commit id fe8a0679
> >
> > I need to finish the work on adding some exact steps or possible docker
> > image to regenerate capabilities.
>
> But aren't some of the responses better tied to native hardware rather
> than virtualized (which I'm assuming a docker image may provide)?
> Especially the CPU ones and as seen from the s390x patch on list the
> response from qemuMonitorJSONGetKVMState.
>
> >
> > I would create a rule that if someone is updating replies files they
> > should check whether the CPU changes are only because of different host
> > CPU or whether there is some actual change. In the first case they
> > should not be part of the patch, it just pollutes the diff with
> > unnecessary changes.
>
> If we had/used the same, native box for each iteration of the
> capabilities files, then sure this probably would be easier unless
> someone (other than me ;-)) wants to mock a "fully featured" response.
>
> What we don't want to do is commit the emulated results, which I believe
> was done in the last go around - check the s390x results for the
> GetKVMState reply (libvirt-7) compared to what's posted upstream from
> real hardware yesterday by Shalini.
>
> I also don't think it's "right" to edit whatever CPU response one
gets
> just because it's run on different hardware just so we can avoid
> polluting the diff. Again, back to my previous points - same, native
> hardware or better mocking...
>
> >
> > Additional note related to the first paragraph, I don't thing that we
> > need to include "xen" related things in replies since they are not
used
> > by QEMU driver.
> >
>
> I was wondering why those xen* things showed up - perhaps because my
> environment included the xen-devel package... After removing, yep those
> go away... I could post another version, but it sounds like this isn't
> desired anyway - so I won't pursue it too much. Just figured it would
> be better to have 2.12 instead of 2.11.90 in the caps/replies that we
> store. The details of CPU differences seem to always be a given. We
> don't "filter" the initial posting based on it so requiring any
update
> posted by someone using different hardware would seem to be overkill
> especially since it doesn't really seem to matter.
>
> John
>
> BTW: It also seems spice* type flags/bits only show up in certain
> environments - so that perhaps too could be considered for some sort of
> mocking. Something that's different between the s390x results posted
> upstream as well as the x86_64 results I created.
Isn't the spice capability depending on how the qemu binary was build?
Yes, when running QEMU configure script you can disable/enable spice
using --disable-spice or --enable-spice. If you don't specify anything
about spice the configure script will use pkg-config to figure out
whether you have spice developer libs installed in your system or not.
Pavel