Re: [Qemu-devel] [PATCH v2] qom: Implement qom-get HMP command

["Daniel P. Berrange" <berrange@redhat.com>]

On Mon, Sep 19, 2016 at 11:18:05AM +0200, Markus Armbruster wrote:
> "Daniel P. Berrange" <berrange@redhat.com> writes:
> 
> > On Wed, Sep 14, 2016 at 11:30:06AM +0100, Dr. David Alan Gilbert wrote:
> >> * Markus Armbruster (armbru@redhat.com) wrote:
> >> > "Dr. David Alan Gilbert (git)" <dgilbert@redhat.com> writes:
> >> > 
> >> > > From: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
> >> > >
> >> > > This started off as Andreas Färber's implementation from
> >> > > March 2015, but after feedback from Paolo morphed into
> >> > > using the json output which handles structs reasonably.
> >> > >
> >> > > Use with qom-list to find the members of an object.
> >> > >
> >> > > (qemu) qom-get /backend/console[0]/device/vga.rom[0] size
> >> > > 65536
> >> > > (qemu) qom-get /machine smm
> >> > > "auto"
> >> > > (qemu) qom-get /machine rtc-time
> >> > > {
> >> > >     "tm_year": 116,
> >> > >     "tm_sec": 0,
> >> > >     "tm_hour": 9,
> >> > >     "tm_min": 46,
> >> > >     "tm_mon": 8,
> >> > >     "tm_mday": 6
> >> > > }
> >> > > (qemu) qom-get /machine frob
> >> > > Property '.frob' not found
> >> > >
> >> > > Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
> >> > 
> >> > Ignorant question: how does qom-set deal with structs?
> >> > 
> >> > I tried the obvious
> >> > 
> >> >     (qemu) qom-set /machine rtc-time abc
> >> >     Insufficient permission to perform this operation
> >> 
> >> I don't think it does.
> >
> > Indeed it can't - qom_set ends up calling object_property_parse which
> > uses string-input-visitor to parse the value, which can only handle
> > scalars as the magic special case list-of-ints.
> >
> > To deal with compound properties would really require us to use a
> > qdict_crumple + qmp_input_visitor combination, similar to how I've
> > made -object and object_add be able to deal with compound properties.
> 
> HMP I/O formats are not ABI.  We can use visitors in whatever way we
> want, as long as we keep -get and -set consistent.  The sane way to do
> that is using the same kind of visitor for both, in its input and output
> form, respectively.
> 
> Right now, qom-set uses the string input visitor.  As long as it does
> that, qom-get should use the string output visitor.  Sadly, this pair of
> visitors is quite limited ("does not implement support for visiting QAPI
> structs, alternates, null, or arbitrary QTypes").  We can extend it to
> cover more, or we can switch to another, less limited pair of visitors.
> 
> Can we agree on what to do so we can have qom-get sooner rather than
> later?  It doesn't have to be perfect, we can iterate.

I think that -object sets the precedent that the rest should ultimately
follow. It currently uses the opts visitor syntax, but is being switched
over to the combination of qdict_crumple + qobject input visitor, which
is basically the same as opts visitor syntax for scalars and with dotted
notation for compound types.

The HMP object_add command will use the exact same syntax as -object
CLI arg. Given this, I think 'qom-set' really ought to be updated to use
qdict_crumple + qobject input visitor too, so it can deal with compound
types. In fact I'd view the lack of conversion of qom-set as a mistake
in my patch series - I should have converted that too, while adding
support for compound properties to -object and object_add.

This ultimately means that qom-get probably ought to use qobject output
visitor, followed by a qdict flatten  operation to turn the nested
dicts/lists, into a flat dict with dotted syntax.

Regards,
Daniel
-- 
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|