I prefer #2, as I do not see the single error to a collection of requests as an issue. The same out of procedure error issue can happen with a request that is malformed or non-existent. However I do feel that the ID is important during multi call in case the client does not remember the order in which they came, something about the blind trust of order bugs me. Not a point I would press however...
#3 works and has been suggested here before. It does make the spec simpler, but it would also remove an implied part of the original spec (IIRC, or I am possibly remembering it wrong). Being compatible with xml-rpc is nice, but I am not sure that means it is the correct way forward.
Just a topic I would want to see mentioned/acknowledged in the spec, #2 or #3 is fine with me at this point if we are making the 1.1WD cleaner.
--
Matthew P. C. Morley
.
