Re: [Qemu-devel] [PATCH 1/8] qapi: fix NULL pointer dereference

[Anthony Liguori]

On 12/16/2011 09:13 AM, Paolo Bonzini wrote:
> On 12/16/2011 04:05 PM, Anthony Liguori wrote:
> > > > What are the uses of null in qdev string properties? I know you can't
> > > > set a string to null since parse() doesn't have a null syntax. So we're
> > > > really just talking about an uninitialized state, right?
> > >
> > > Yes. No ROM BAR is an example of a NULL string property.
> >
> > So it's really filenames and backend names, right? Can we just treat
> > empty strings like NULL? I think all of the various bits handles both
> > the same way.
>
> The only case I can think of when it differs is serial numbers. An empty serial
> number is different from no serial number (which means do not expose one at 
> all).
>
> However, DEFINE_PROP_STRING does not allow setting a default, and some device
> models rely on a NULL value as a default. I see two ways out: 1) add nullable
> strings; 2) merge without this patch, knowing qom-get is kind-of broken, and
> proceed adding a default to all DEFINE_PROP_STRING; this would be a merge
> nightmare for you.

I thrive on pain it seems :-)

> 3) merge without this patch, knowing it's kind-of broken, and
> later decide what to do.

Ok.  I think nullable strings are not a good idea simply because it means that a 
property can have a state that cannot be set.

Long term, I want to be able to do something like dump the current device graph 
to a config file, and then use that config file to recreate the same machine 
again.  A nullable property without a null representation would not allow this.

Regards,

Anthony Liguori

> Let's do (3), and add nullable strings if the discussion stalls.
>
> Paolo