On 06/12/12 20:48, Martynas Jusevičius wrote:
Andy, we've been through this already:
http://mail-archives.eu.apache.org/mod_mbox/jena-users/201202.mbox/%3ccae35vmzpcczxswt37etsb_5bzw6qgv-uufphy8bx6ssh4u9...@mail.gmail.com%3E
Have you looked at Stephen's work on jena-client?
OK, maybe I should say JAX-RS and Jersey as its implementation, which
also includes the client.
The question is about what's on the wire.
Dydra returns HTML when no content was asked for. It was a POST with
the request content type as application/sparql-update. And no "Accept"
because there is no wanted reply.
Why not 204?
And why is a SPARQL query result set returned for */*?
JAX-RS allows not only high-level annotations, but also low-level
conneg if there is a need for it, both on server and client sides.
Here's an example of actual code I have at hand:
ResultSet resultSet = getResultSet();
EntityTag entityTag = new
EntityTag(Long.toHexString(ResultSetUtils.hashResultSet(resultSet)));
Response.ResponseBuilder rb =
getRequest().evaluatePreconditions(entityTag);
if (rb != null)
{
if (log.isTraceEnabled()) log.trace("Resource not modified,
skipping
Response generation");
return rb.build();
}
else
{
Variant variant =
getRequest().selectVariant(RESULT_SET_VARIANTS);
if (variant == null)
{
if (log.isTraceEnabled()) log.trace("Requested Variant {}
is not
on the list of acceptable Response Variants: {}", variant,
getVariants());
return Response.notAcceptable(RESULT_SET_VARIANTS).build();
}
else
{
if (log.isTraceEnabled()) log.trace("Generating SPARQL
results
Response with Variant: {} and EntityTag: {}", variant, entityTag);
return Response.ok(resultSet,
variant).tag(entityTag).build();
// uses ResultSetWriter
}
}
That is the server side of a query operation.
The question is about the client side of an update operation.
The point about using a standard (SPARQL Protocol) is that the client
library (ARQ) can provide one single call operation to send a SPARQL
Update to the server. It's higher level than exposing HTTP details.
Jena has different HTTP code scattered all over the place,
Really?
HttpQuery is the only mention of HttpURLConnection in ARQ I can find.
All use in jena-core is to be replace by the new RIOT reader (and is
already if you have ARQ initialized).
For SPARQL Update and model reading, ARQ uses Apache Common HttpClient.
This is in 2 files and soon to be 1 -- HttpOp and the new material to
be merged from HttpOp2.
while
JAX-RS/Jersey covers all HTTP use cases (that I came across) within
Linked Data -- what else do we need?
Encapsulation of low level details.
Andy
Martynas
On Thu, Dec 6, 2012 at 10:36 PM, Rob Vesse <[email protected]> wrote:
On 12/6/12 12:31 PM, "Andy Seaborne" <[email protected]> wrote:
On 06/12/12 20:03, Martynas Jusevičius wrote:
If Jena would be consistent in its HTTP handling and use an
established API such as JAX-RS, things could be much easier.
Could you explain that a bit more so we can see what can be improved (or
submit a patch)? Do you have specific suggestions?
JAX-RS is an API for mapping Java with annotations to web resources.
HTTP is a protocol.
I agree, I don't see how JAX-RS is relevant here.
However perhaps we should be ensuring that everywhere that does anything
HTTP uses the Apache HttpClient as a point of standardization and provide
ways to better expose that up to users to add custom configuration?
As Andy says patches are welcome in this area. I know I've done some work
in the past to improve configurability on the remote query side but it
still uses the java.net APIs rather than HttpClient as the remote update
stuff does.
The question is about the use of content negotiation on SPARQL update
requests i.e. the bits on the wire. Surely the client API or server
implementation should work with any legal bits on the wire (and be
defensive about illegal bits?)
+Many
Though we should consider Postel's law here - "Be liberal in what you
accept and conservative in what you send"
In this particular case we are complying with this and Dydra is not as I
see it.
Rob
Andy
Martynas
graphity.org
On Thu, Dec 6, 2012 at 6:04 PM, Rob Vesse <[email protected]> wrote:
My two cents:
In my experience of writing a HTTP connector for Dydra for dotNetRDF it
proved to be a pretty awful implementation of HTTP conneg and SPARQL
protocols in general compared to other similar systems like Sesame,
Stardog, Fuseki etc. and I had to put an awful lot of implementation
specific hacks in place to get things working.
For example I have a still open bug related to something very similar
to
this
(https://getsatisfaction.com/dydra/topics/basic_digest_http_authenticati
on_
is_not_supported) which relates to them doing dumb stuff with conneg
and
not just using HTTP authentication nicely. In essence if you are not
logged in and the Accept header does not ask for machine readable
formats
they automatically redirect to a HTML login form instead of doing HTTP
auth. This is particularly problematic because their processing of the
Accept header is poor, if you so much as mention HTML in there it will
think you are a browser even if machine readable formats are ahead of
HTML
in the Accept header.
I agree with Andy that maybe there is some room for improvement in
Jena to
allow this to be configured in general but I wouldn't want to turn
this on
in general just to work with Dydra which I'm still expecting to die a
death. I know for a fact that one of their core developers has long
since
parted ways with the company and I haven't seen any activity on any of
my
various open bugs in a long time.
Rob
On 12/6/12 3:14 AM, "Andy Seaborne" <[email protected]> wrote:
Hi Simon,
What a nuisance. It's not, as far as I know, illegal, but it is a
rather odd interpretation of HTTP POST for remote operations.
Not sending "Accept" is because it's a POST which does not need to
return anything. conneg for an HTML page is find but it's not conneg
if
it returns it without being asked for! I'm a bit worried that
changing
in the general casefor Dydra means it will be inconvenient elsewhere.
(And don't some (old) browsers send */* always?)
So it's will need some sort of configuration mechanism to make it
endpoint specific.
Good news - every UpdateProcessor has a Context object that is
associated with the request. It's currently null for a remote request
but the mechanism is at least there.
Remote query (either QueryExecutionFactory.sparelService or SERVICE
in a
query) has a mechanism for setting the request with custom headers.
See ARQ.serviceParams and class Service.
Do you want to raise a JIRA for this? And (ideally) contribute a
patch?
Andy
On 05/12/12 21:43, Simon Gábor wrote:
Hi Andy,
I am trying to use Dydra (http://dydra.com/). It requires to send
Accept
header, otherwise it returns with a full HTML webpage (with Dydra's
online
query tool on it). For update commands it is happy with just an
Accept:
*/*
then returns 200 and a simple boolean SPARQL XML result.
Example:
POST http://dydra.com/orkszoft/test01/sparql HTTP/1.1
Accept: */*
Content-Length: 364
Content-Type: application/sparql-update
Host: dydra.com
Connection: Keep-Alive
User-Agent: Apache-HttpClient/4.2.2 (java 1.5)
Authorization: Basic ***
PREFIX dc: <http://purl.org/dc/elements/1.1/>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>
PREFIX owl: <http://www.w3.org/2002/07/owl#>
PREFIX xsd: <http://www.w3.org/2001/XMLSchema#>
PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
DELETE DATA {
<http://dbpedia.org/resource/San_Francisco> rdfs:label "My San
Francisco"
.
}
Response:
HTTP/1.1 200 OK
Content-Type: */*
Connection: keep-alive
Status: 200
X-Powered-By: Phusion Passenger (mod_rails/mod_rack) 3.0.9
Cache-Control: no-cache
X-UA-Compatible: IE=Edge,chrome=1
Set-Cookie: _dydra_session=****; path=/; HttpOnly
X-Runtime: 0.849964
Server: nginx/1.0.8 + Phusion Passenger 3.0.9 (mod_rails/mod_rack)
Access-Control-Allow-Origin: *
Access-Control-Allow-Credentials: *
Content-Length: 123
<?xml version='1.0'?>
<sparql xmlns='http://www.w3.org/2005/sparql-results#'>
<head/>
<boolean>true</boolean>
</sparql>
I'm having trouble to decide whether Dydra has a faulty design (it's
not
conform with the protocol recommendation) or Jena has no support for
the
optional content negotiation (or both)?
It's for sure that Dydra's method is somewhat weird because for a
wrong
SPARQL it returns code 400 - so it seems that the response body
holds no
new info.
Gabor Simon
2012/12/5 Andy Seaborne <[email protected]>
On 03/12/12 12:44, Simon Gábor wrote:
Hi All,
According to the SPARQL Update protocol recommendation:
"The response body of a successful update request is implementation
defined. Implementations *may* use HTTP content negotiation to
provide
both
human-readable and machine-processable information about the
completed
update request."
How can i specify accept header in the SPARQL Update request? I
cannot
find
a way in UpdateProcessor (UpdateProcessRemote) or in UpdateRequest.
Simon,
There currently isn't a way to do that and also the .execute()
operation
does not return anything. Errors appear as exceptions driven from
the
HTTP
response code.
What had you in mind?
I don't know of any systems currently that return different entity
bodies
based on conneg. Are there any?
The reason in the spec for "MAY" is that there is no standard format
for a
reply. RDFa, RDF, JSON [1] (why have an RDF processor when you send
a
string?) all make sense in different scenarios.
Andy
[1]
http://tools.ietf.org/html/**draft-nottingham-http-problem-**01<http:
//t
ools.ietf.org/html/draft-nottingham-http-problem-01>
