Re: [Qemu-devel] [RFC] New Migration Protocol using Visitor Interface

[Anthony Liguori <anthony@codemonkey.ws>]

On 10/03/2011 08:30 AM, Michael S. Tsirkin wrote:
> On Mon, Oct 03, 2011 at 08:18:31AM -0500, Anthony Liguori wrote:
>> On 10/03/2011 08:10 AM, Stefan Berger wrote:
>>> I am doing that. Indefinite length encoding *would* be a problem because you
>>> cannot push the size onto the stack so that you could skip to the end of the
>>> structure.
>>
>> For an indefinite length encoding, you just have to keep reading the
>> stream at end_struct until you hit the canary tag ignoring anything
>> you encounter.
>>
>> Regards,
>>
>> Anthony Liguori
>
> That's not exactly right: one indefinite length encoding can be nested
> within the other. So what we must do is keep reading, read out,
> decode and skip regular length encodings, and count the
> nesting of indefinite length encodings. When we see bit 7 set,
> we increase nesting. When we see end of content, we descrease nesting.
> Stop when nesting reaches 0.

Yes, you basically have a skip field in the Visitor.  If skip is set, then don't 
actually marshal to anything.  So:

type_int (et al.)
   if (!skip) {
      store to passed in int
   }
   increment offset

Then when you want to skip to the end of the indefinite, you do something like:

skip_indefinite:
   while tag != CANARY:
     if tag == INT:
       visit_type_int(v, NULL, NULL, errp);
     elif tag == STRING:
       visit_type_str(v, NULL, NULL, errp);
     elif tag == INDEFINITE:
       visit_start_struct(v, NULL, NULL, errp);
       skip_indefinite(v, errp);
       visit_end_struct(v, NULL, NULL, errp);

end_struct:
   v->skip = true;
   skip_indefinite(v, errp);
   v->skip = false;

Regards,

Anthony Liguori

>