Re: [Qemu-devel] Re: [RFC] qapi: events in QMP

[Luiz Capitulino]

On Mon, 14 Feb 2011 13:34:11 -0600
Anthony Liguori <address@hidden> wrote:

> On 02/14/2011 12:34 PM, Luiz Capitulino wrote:
> > On Mon, 14 Feb 2011 08:39:11 -0600
> > Anthony Liguori<address@hidden>  wrote:
> >
> >    
> >> On 02/14/2011 06:45 AM, Luiz Capitulino wrote:
> >>      
> >>> So the question is: how does the schema based design support extending
> >>> commands or events? Does it require adding new commands/events?
> >>>
> >>>        
> >> Well, let me ask you, how do we do that today?
> >>
> >> Let's say that I want to add a new parameter to the `change' function so
> >> that I can include a salt parameter as part of the password.
> >>
> >> The way we'd do this today is by checking for the 'salt' parameter in
> >> qdict, and if it's not present, use a random salt or something like that.
> >>      
> > You likely want to do what you did before. Of course that you have to
> > consider if what you're doing is extending an existing command or badly
> > overloading it (like change is today), in this case you'll want to add
> > a new command instead.
> >
> > But yes, the use-case here is extending an existing command.
> >
> >    
> >> However, if I'm a QMP client, how can I tell whether you're going to
> >> ignore my salt parameter or actually use it?  Nothing in QMP tells me
> >> this today.  If I set the salt parameter in the `change' command, I'll
> >> just get a success message.
> >>      
> > I'm sorry?
> >
> > { "execute": "change", "arguments": { "device": "vnc", "target": 
> > "password", "arg": "1234", "salt": "r1" } }
> > {"error": {"class": "InvalidParameter", "desc": "Invalid parameter 'salt'", 
> > "data": {"name": "salt"}}}
> >    
> 
> So I'm supposed to execute the command, and if execution fails, drop the 
> new parameter?  If we add a few optional parameters, does that mean I 
> have to try every possible combination of parameters?

No, of course not, our plan has always been to do this via an schema,
the only reason we don't do this today is lack of time/help.

> >> Even if we expose a schema, but leave things as-is, having to parse the
> >> schema as part of a function call is pretty horrible,
> >>      
> > That's a client implementation detail, they are not required to do it
> > as part of a function call.
> >
> > But let me ask, if we don't expose a schema, how will clients be able to
> > query available commands/events and their parameters?
> >    
> 
> We need to expose the schema, I'm not saying we shouldn't.  But we don't 
> today.
> 
> You're arguing that we should extend commands by adding new parameters.  

Commands and events, you haven't commented on events yet and that seems
a bit worse than commands.

> I'm saying that's a bad interface.  If we need to change a command, we 
> should introduce a new command.  It's a well understood mechanism for 
> maintaining compatibility (just about every C library does exactly this).

So, let's agree we disagree.

> >> particularly if
> >> distros do silly things like backport some optional parameters and not
> >> others.  If those optional parameters are deeply nested in a structure,
> >> it's even worse.
> >>      
> > Why would they do this? I mean, if distros (or anyone else shipping qemu)
> > goes that deep on changing the wire protocol they are on their own, why
> > would we want to solve this problem?
> >    
> 
> It's not at all unreasonable for a distro to backport a new QMP 
> command.  If all modifications are discrete commands, compatibility is 
> easy to preserve, however if a distro does backporting and we end up 
> with a frankenstein command, compatibility will be an issue.

I disagree. Let's say we have added three new arguments to the command foo,
and now we have foo1, foo2 and foo3. I'm a quite old distro and only
have foo, which command should I backport? All of them? Only the latest?

I can't see how easier this is. Backporting APIs will almost always suck.

> >> OTOH, if we introduce a new command to set the password with a salt, it
> >> becomes very easy for the client to support.  The do something as simple 
> >> as:
> >>
> >> if qmp.has_command("vnc-set-password-with-salt"):
> >>       qmp.vnc_set_password_with_salt('foobar', 'X*')
> >> else:
> >>       window.set_weak_security_icon(True)
> >>       qmp.vnc_set_password('foobar')
> >>
> >> Now you could answer, hey, we can add capabilities then those
> >> capabilities can quickly get out of hand.
> >>      
> > Adding one command per new argument has its problems too and it's even
> > worse with events, as clients will have to be changed to handle a
> > new event just because of a parameter addition.
> >    
> 
> Yes, but it's an extremely well understood way to design compatible APIs.

For C, yes. But one of the main goals of a high level protocol is to be
language independent, isn't it?

> > Look, although I did _not_ check any code yet, your description of the QAPI
> > looks really exciting. I'm not against it, what bothers me though is this
> > number of small limitations we're imposing to the wire protocol.
> >
> > Why don't we make libqmp internal only? This way we're free to change it
> > whatever we want.
> >    
> 
> libqmp is a test of how easy it is to use QMP from an external 
> application.  If we can't keep libqmp stable, then that means tools like 
> libvirt will always have a hard time using QMP.
> 
> Proper C support is important.  We cannot make it impossible to write a 
> useful C client API.

I wouldn't say it's impossible, but anyway, the important point here is
that we disagree about the side effects QAPI is going to introduce in QMP,
I don't know how to solve this, maybe we can discuss this upstream, but I'm
not sure the situation will change much.

Should we vote? :)