From: "Theo van Veen" <[log in to unmask]>
> I see, but in this case the client has to have pre-knowledge about the
> server. I consider interoperability as minimizing the need for
> pre-knowledge. In this case I would do it the opposite way: the server
> doesn't send the numberOfRecords and the client can request by extension
> "please give me the numberOfRecords anyway" rather than let the client
> say "I don't care about the numberOfRecords so don't send a diagnostic".
There are many possible approaches if we were not constrained by the need
for compatibility with version 1.x.
The server not sending the count, i.e. making the parameter optional, is
part of the discussion for version 2.0 - it would break version 1.x.
I ask that we defer 2.0 discussion for now (for brief time until we can set
up procedures for maintaining these parallel discussions without getting
confused).
It seems to me that within the constraints of version 1.x the best we can do
is the solution we've come up with - provide the client with a means to tell
the server not to bother counting the records (and the only such means
within 1.x would be an extension), and to define a diagnostic or two (or so)
that says that the count is approximate, a wild guess, a filler value, etc.
--Ray
|