Re: [PATCH 0/6] [VERY RFC] Migration Stream v2

[George Dunlap <george.dunlap@eu.citrix.com>]

On 04/15/2014 11:35 AM, Ian Jackson wrote:
> David Vrabel writes ("Re: [Xen-devel] [PATCH 0/6] [VERY RFC] Migration Stream v2"):
>> On 14/04/14 18:49, George Dunlap wrote:
>>> What did you imagine might constitute an "Optional" record?
>>
>> This was something Ian Jackson asked for and it seems like a useful
>> capabilitity to have for future use.  Not sure what it might be used for
>> yet.
>
> Right.
>
> Long experience with protocol design has taught me that protocols
> should almost always have both an extensibility mechanism which is
> ignored by ignorant receivers, and one which causes ignorant receivers
> to abort.
>
> I don't know yet what we might use it for.  However, we should test
> that it works (ie is ignored by) the receiver (or it will be useless).

Yes, this is the main concern.  2 billion record types should be plenty 
for the "required" field, so the 2 billion allocated for "optional" 
shouldn't be a big loss. :-)  The main risk would be if something which 
is, in fact, required for proper operation on the far side is marked 
"optional".  I guess as long as we have an "ignore everything optional" 
test case we should be OK.

  -George