On Wed, Jun 22, 2011 at 04:48:31PM +0100, Daniel P. Berrange wrote:
On Wed, Jun 22, 2011 at 09:41:57AM -0600, Eric Blake wrote:
> On 06/21/2011 09:53 AM, Matthias Bolte wrote:
> >> [1]
http://libvirt.org/html/libvirt-libvirt.html#virConnectClose
> >>
> >
> > In contrast to the documentation virConnectClose returns the remaining
> > reference count of the connection after unref'ing it. This means
> > virConnectClose returns -1 on error and returns 0 when there is no
> > reference left and the connection is really closed. When it returns >
> > 0 then the connection isn't closed yet, but is kept open because there
> > are still objects alive that depend on it. For example when you open a
> > connection and get a virDomainPtr from it then the refcount of the
> > connection is increased, because the domain object depends on it. When
> > the domain object is freed then the refcount of the connection is
> > decreased again.
> >
> > All objects in libvirt like connections, domains, networks etc are
> > reference counted. But in contrast to virConnectClose other free
> > functions like virDomainFree don't return the remaining reference
> > count, they really just return -1 or 0. So the question is: do we
> > change virConnectClose to really just return -1 or 0, or do we update
> > the documentation, because someone might depend on virConnectClose
> > allowing to detect via its return value whether the connection was
> > really closed or not.
>
> At this point, I worry that changing return values might break existing
> code, so I would favor a documentation update, but I'd also like to hear
> an opinion from DV or danpb.
I think we shouldn't change the semantics.
Just to clarify: The code is right. The docs are wrong.
This is actually a good general rule for resolving any doubts :-)
Daniel
--
|: