HTTP content negotiation is about the content-type of the entire
response. That is different from SRU 1.x's recordPacking parameter,
which is about the type of records included IN the response. So we
can't use the former in place of the latter.
Something along the lines of Ralph's suggestion does less violence to
the model -- though it's still not really right, as the parameter's
new and more explicit name in SRU 2.x indicates.
Seems to me that the right thing is to leave SRU 1.x as it is, since
it's been superseded, and to find a non-hackish way to add JSON
record-content to SRU 2.x.
-- Mike.
On 18 February 2014 08:53, Jörg Prante <[log in to unmask]> wrote:
> From what I understand, "recordPacking" was trying to carry information
> of what Z39.50 OIDs did in 1.2.840.10003.5.109.x
> http://www.loc.gov/z3950/agency/defns/oids.html
>
> But today, is there any SRU service which is not working over HTTP (I
> do not mean Z39.50 over SRU)?
>
> There is the MIME type application/sru+xml
> https://tools.ietf.org/html/rfc6207 which is also meant for use in
> the HTTP header "Accept" for content negotation.
>
> I think it's time for SRU / SearchRetrieve to fully embrace HTTP
> content negotiation, and REST - I mean, most importantly, the HTTP
> status codes. Therefore, I would welcome deprecating the parameter
> "recordPacking", and also "recordSchema".
>
> If content negotiation is not sufficient, it could be an option to use
> HTTP headers for additional hinting, for XML schemas, bibliographic
> field formats etc. HTTP headers could also carry Z39.50 OIDs in their
> full extent, by preceding the header name with something like
> "X-Z3950-...", or optional SRU parameters, in "X-SRU-..."
>
> Here's my proposal. This is an SRU request for XML
>
> GET /sru/...
> Host: localhost
> Accept: application/sru+xml
> ->
> HTTP/1.1 200 OK
> Content-Type: application/sru+xml
>
> and this is for JSON
>
> GET /sru/...
> Host: localhost
> Accept: application/sru+json
> ->
> HTTP/1.1 200 OK
> Content-Type: application/sru+json
>
> or even, with HTTP content negotiation, both are possible:
>
> GET /sru/...
> Host: localhost
> Accept: application/sru+xml,application/sru+json
> ->
> HTTP/1.1 200 OK
> Content-Type: application/sru+xml
>
> And also something for RDF serializations, because Bibframe is coming
> close.
>
> I have only JSON-LD, so over the SRU endpoint, I would also like to
> perform RESTful actions on RDF graphs, like
>
> GET /sru/...
> Host: localhost
> Accept: application/ld+json
> ->
> HTTP/1.1 200 OK
> Content-Type: application/ld+json
>
> Such requests lack any information about "recordSchema",
> "recordPacking", OIDs, Z39.50, simply because there is none such
> information. If it was, it's all encoded in the JSON-LD.
>
> HTTP GET requests are stateless. So I'd like to drop SRU parameter
> "resultSetTTL". Was this ever used, clients enforcing servers to keep
> result sets, for days or weeks, or even longer? If so, it could be
> replaced by URL- or cookie-based session life times, under control of
> the server.
>
> Best,
>
> Jörg
>
>
>>>> "LeVan,Ralph" <[log in to unmask]> schrieb am 17.02.2014 um 21.27 Uhr
> in Nachricht
> <[log in to unmask]>:
>> I'm playing around with JSON and SRU. My server has pretty decent
> support
>> for content negotiation and I've started explaining to it how to
> handle
>> requests for JSON.
>>
>>
> http://rdap02pxdu.dev.oclc.org:8080/viaf/search?query=local.names+all+%22ral
>> ph%20levan%22&httpAccept=application/json
>>
>> That's a simple name search, with a request that the response be in
> JSON.
>> You'll notice that the recordPacking is "xml", which is appropriate,
> because
>> the records are returned as XML. But, as far as JSON is concerned,
> what I
>> returned was a string. (The value of recordData in JSON is of type
> string.)
>> What happens if I want to return JSON records? How do I even ask for
> JSON
>> records?
>>
>> Right now, to return JSON records I have to say that the
> recordPacking is
>> "string" (because it sure isn't "xml") and pick a recordSchema of
> JSON
>> (because there's no other place to say you want JSON.
>>
> http://rdap02pxdu.dev.oclc.org:8080/viaf/search?query=local.names+all+%22ral
>>
> ph%20levan%22&recordPacking=string&recordSchema=JSON&httpAccept=application/json
>>
>>
>> I think the recordPacking was intended to be the mimeType of the
> record, as
>> opposed to the httpAccept, which is the mimeType of the response. I
> think
>> the recordSchema was intended to talk about the logical content of
> the
>> record, not the physical format. In my particular case, I already
> have
>> multiple schemas that I want to return my data in (LinkedData,
> Marc21,
>> VIAFCluster), regardless of the mimeType that the record is returned
> in.
>>
>> My suggestion is that we expand the meaning of recordPacking to mean
>
>> mimeType of the recordData and grandfather in the special values of
> "string"
>> to mean "text/plain" and "xml" to mean "text/xml". If you return
> text/plain
>> records in a text/xml response, it is up to the application to make
> the
>> records safe in the XML (meaning escaping the appropriate characters
> in the
>> string.) That same would be expected in JSON (where the string would
> just
>> get quotes around it.)
>>
>> Does this seem sensible? Is anyone else doing anything like this?
>>
>> Thanks!
>>
>> Ralph
>
>
>
>
> --
> Jörg Prante
> hbz, Gruppe Portale
> - Digitale Bibliothek und Online-Fernleihe -
> Postfach 270451, 50510 Köln, Deutschland
> Telefon +49-221-40075-156, Fax +49-221-40075-190
> [log in to unmask]
> http://www.hbz-nrw.de
>
|