On 12/11/2015 04:10 AM, Paolo Bonzini wrote:
> 
> 
> On 11/12/2015 00:53, Eric Blake wrote:
>> Instead of rolling our own limited JSON outputter, we can just
>> wrap the more full-featured JSON output Visitor.
>>
>> This slightly changes the output (different spacing), but the
>> result is still equivalent JSON contents.
>>
>> Signed-off-by: Eric Blake <eblake@redhat.com>
>> ---
>>  qjson.c | 61 ++++++++++++++++++++++---------------------------------------
>>  1 file changed, 22 insertions(+), 39 deletions(-)
> 
> I didn't find out which tree your patches apply to, but this:

Mentioned in the cover letter:

+ v8 or later of my qapi subset E patches (at the time of this mail,
only v7 has been posted):
https://lists.gnu.org/archive/html/qemu-devel/2015-12/msg01181.html

Also available as a tag at this location (with prerequisites applied):
git fetch git://repo.or.cz/qemu/ericb.git qapi-jsonv1

(note that it is only a tag, not a branch, and you are right that it
doesn't yet apply to any maintainer's tree).


> @@ -1068,9 +1070,11 @@ void qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only)
>          return;
>      }
>  
> -    vmdesc = qjson_new();
> -    json_prop_int(vmdesc, "page_size", TARGET_PAGE_SIZE);
> -    json_start_array(vmdesc, "devices");
> +    vmdesc_jov = json_output_visitor_new();
> +    vmdesc = json_output_get_visitor(vmdesc_jov);
> +    visit_start_struct(vmdesc, NULL, 0, NULL, &error_abort);
> +    visit_type_int(vmdesc, TARGET_PAGE_SIZE, "page_size", &error_abort);
> +    visit_start_list(vmdesc, NULL, 0, "devices", &error_abort);
>      QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {

Nice idea.

> compiles and is pretty much a 1:1 translation from the qjson.c API to the
> visitor API (using this patch as a guide).  Feel free to include it and
> remove qjson.c.  Alternatively, you can leave out this patch and I'll test
> and submit the transition.

Should I squash the two together, or keep my current patch and drop
qjson.c as a separate patch?

Also, while reading this, I noticed that qjson.h has signatures that
match JSON ('name' comes before 'value', when generating a "name":value
pair), while visitor.h has signatures that are backwards ('obj' comes
before 'name', even though the output will be "name":value-of-obj).  I'm
seriously debating about a tree-wide coccinelle patch to reverse the
order or arguments of all the visit_type_* functions.

-- 
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org