On Mon, Mar 31, 2008 at 01:02:49PM +0100, Daniel P. Berrange wrote:
On Mon, Mar 31, 2008 at 11:00:56AM +0100, Richard W.M. Jones wrote:
> The 'conn', 'dom' and 'net' fields in a virterror can't
be accessed
> safely, particularly from a garbage-collected language. The only safe
> thing one could do with them is a physical equality test on the
> pointer against an existing conn/dom/net pointer.
>
> But the solution isn't to copy the dom object, since in a GC-d
> language someone could grab a handle to the copied dom which would
> have a different lifetime from the error object. If anything the
> solution would be to remove those fields from the error object since
> they are highly unlikely to be useful in any real world application.
> We can't do this because of backwards compatibility promises but we
> should instead deprecate them.
We should at the very least NULL-ify the dom/net fields in the Error
object associated with the Connection when we free the Domain/Network
object.
agreed, very simple test but avoids dandling pointers.
We also probably need to re-arrange the remote daemon code a
little so that it serializes to XDR format before the Domain/Network object
are free'd.
I agree that deep copying isn't the answer here & messes up the reference
counting and will cause potential memory leaks.
yes, this will become unmanageable very fast i'm afraid.
Daniel
--
Red Hat Virtualization group
http://redhat.com/virtualization/
Daniel Veillard | virtualization library
http://libvirt.org/
veillard(a)redhat.com | libxml GNOME XML XSLT toolkit
http://xmlsoft.org/
http://veillard.com/ | Rpmfind RPM search engine
http://rpmfind.net/