----- Original Message -----
> On Wed, Sep 21, 2016 at 02:36:28PM +0400, Marc-André Lureau wrote:
> > Some getters already set *obj argument to NULL early, let's do this for
> > all for consistent behaviour in case of errors.
> > 
> > Signed-off-by: Marc-André Lureau <marcandre.lur...@redhat.com>
> If we want consistent behaviour, there's plenty more visit methods
> that need updating beyond these two.  eg input_type_int64 will
> leave '*obj' untouched on error.
> In fact if we want to have '*obj' given a NULL value on error,
> then it seems we should instead add code to 'qapi-visit-core.c'
> to always initialize '*obj' to NULL, instead of doing it in
> qmp-input-visitor.c That way all visitor implementations get
> the same behaviour

I think that's not easily doable, as an input visitor will want to set *obj (to 
NULL or something), but the output visitor may need *obj != NULL as an input.

It'snot really elegant that there is visitor-input/output specific code already 
in the visit-core, I would rather have that code in the respective visitors.

> Alternatively, we could say that '*obj' should never be touched
> on error paths, in which case we've a bunchof cleanup todo
> in qmp_input_visitor to avoid splattering *obj.
> Regards,
> Daniel
> --
> |: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
> |: http://libvirt.org              -o-             http://virt-manager.org :|
> |: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
> |: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|

Reply via email to