Re: [Qemu-devel] [PATCH 10/19] monitor: Convert do_info_name() to QObject

[Avi Kivity]

On 12/10/2009 06:03 PM, Anthony Liguori wrote:
> Avi Kivity wrote:
> > On 12/10/2009 02:56 PM, Anthony Liguori wrote:
> > > I'd prefer an empty dict.  I actually prefer that over null.
> >
> > Depends.  Key not present suggests the feature is not implemented.  
> > Value is null suggests the feature is not used.  Both key and value 
> > present suggest the feature is in implemented and active.
>
> What I suggested to Luiz was that all info commands return a 
> dictionary.  The main reason was that for most commands, we can extend 
> the commands in a compatible way by adding new fields to the dictionary.

Oh yes.

> My expectation is that there will be a lot of client code that does:
>
> hpet_info = qmp.info_hpet()
> if hpet_info.has_key('period'):
>   period = hpet_info['period']
> else:
>   period = 100 # old qemu's has a fixed period of 100ns
>
> So in keeping with this idiom, I think that checking 
> name_info.has_key('name') is a more meaningful way to determine 
> whether the virtual machine has been given a name vs comparing 
> name_info['name'] == None.

But we have two null conditions to check for, in an extendible dictionary:

1. The feature is unknown to qemu
2. The feature is known to qemu, but disabled

So, "if 'field' in result:" tests the former, and "if result['field']:" 
tests the latter.

In your example, a period of None makes no sense, so it would be 
sufficient to

  period = hpet_info.get('period', 0.100)



--
error compiling committee.c: too many arguments to function