SRU/CQL: clarification about fatal diagnostics

[Oliver Schonefeld <schonefeld@ids-mannheim.de>]

Hello SRU/CQL working group,

currently, I am writing a Java implementation for a SRU server and a SRU
client (for now only Version 1.1/1.2) and from reading the specs it is
not quite clear to me how the exact XML serialization for fatal
diagnostics should look like.
E.g, if a client tries to perform a "searchRetrieve" operation and omits
the query parameter, shall the fatal diagnostic be wrapped inside a
<searchRetrieveResponse> element, like:
<sru:searchRetrieveResponse xmlns:sru="..">
  <sru:version>..</sru:version>
  <sru:diagnostics>
     <diag:diagnostic xmlns:diag="..">
       ..
     </diag:diagnostic>
  </diagnostics>
</sru:searchRetrieveResponse>

And if so, should a fatal diagnostic for e.g. a "scan" operation than
analogously be wrapped inside a <scanResponse> element?
Or is the server supposed to return the diagnostics element? Or any
other serialization?
Various currently existing servers have a different behavior. E.g, the
Oxford English Dictionary[1] wraps in the <searchRetrieveResponse>
element, while the Library of Congress[2] puts the diagnostic in the
<diagnostics> element of a <explainResponse> and the German National
Library[3] just sends the diagnostics. What is the correct behavior?
The available XML schemas for SRU don't really provide better insights.
Maybe the section in the spec can be made more definite, maybe with an
explicit example of the complete XML serialization of a fatal diagnostic
for a, e.g. "searchRetrieve" operation.

Btw, IMHO the behavior of the Oxford English Dictionary makes most sense.

Thanks and best regards,
 Oliver

[1] http://www.oed.com/srupage
[2] http://z3950.loc.gov:7090/voyager
[3] http://services.dnb.de/sru/dnb
-- 
Oliver Schonefeld
Institut für Deutsche Sprache, Zentrale Forschung
R5, 6-13, D-68161 Mannheim
+49-(0)621-1581-451 | http://www.ids-mannheim.de