On 10/25/2016 08:41 AM, Eric Blake wrote:
>>> +static void test_visitor_out_partial_visit(TestOutputVisitorData *data,
>>> + const void *unused)
>>> +{
>>> + /* Check that aborting mid-visit doesn't leak or double-free */
>>> + visit_start_struct(data->ov, NULL, NULL, 0, &error_abort);
>>> + visit_start_struct(data->ov, "nested", NULL, 0, &error_abort);
>>> + visitor_reset(data);
>>
>> Add a second test that additionally visits a member? Or is visiting the
>> "nested" member enough?
>
> Visiting "nested" was enough to trigger the particular valgrind trace I
> wanted, but more coverage never hurts. I'll send v2...
>
>> Would you be willing to do the same for lists and alternates?
>
> ...and it will also cover these.Aborting in the middle of an alternate is trickier, since the current code for alternates does not allow a non-NULL *obj to visit_start_alternate() (that is, no virtual alternate walk is needed; to output an alternate, just call directly to the branch of interest, without a need to wrap things in visit_start/end_alternate; and on input, an alternate is currently useful only into a QAPI C struct). > >> >> Speaking of alternates, visitor.h's virtual walk example covers struct >> and list, but not alternate. > > Sounds like my v2 will be a longer series :) I'm still trying to see whether tweaking the docs to cover alternates makes much sense; we'll see what the patch looks like, but may decide it was not worth it. -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org
signature.asc
Description: OpenPGP digital signature
