Re: LDPR Interaction Model on Create

Hi Nandana,

On Thu, Oct 9, 2014 at 12:21 PM, Nandana Mihindukulasooriya <
nmihindu@fi.upm.es> wrote:

> Hi,
>
> On Thu, Oct 9, 2014 at 4:09 PM, Andrei Sambra <andrei@w3.org> wrote:
>
>> Hi,
>>
>> On 10/09/2014 09:10 AM, Steve Speicher wrote:
>> > Hi,
>> >
>> > Thanks,
>> > Steve Speicher
>> > http://stevespeicher.me
>> >
>> > On Tue, Oct 7, 2014 at 3:48 PM, James Leigh <james@3roundstones.com>
>> wrote:
>> >
>> >> Hello,
>> >>
>> >> Section 5.2.3.4 (copied below) could use some more explanation. In
>> >> particular the first bullet point is not clear. The example given is
>> >> when the created content contains an rdf:type triple indicating a type
>> >> of LDPC, but specifies a LDPR interaction model.
>> >>
>> >> Given section 5.2.1.1 (each LDPC MUST also be a conforming LDPRS) and
>> >> section 4.3.11 (each LDPRS MUST also be a conforming LDPR), I don't
>> >> understand under what conditions a LDPC could NOT also be a LDPR
>> >> interaction model.
>> >>
>> >> Furthermore given the LDP schema, I would expect a POST to a container
>> >> with a Link:<http://www.w3.org/ns/ldp#Resource>;rel="type" that
>> created
>> >> a LDPC member to be successful, since ldp:Container rdfs:subClassOf+
>> >> ldp:Resource and with RDFS entailment all ldp:Container members are
>> also
>> >> ldp:Resource members.
>> >>
>> >
>> > Perhaps it could be clarified that specifying the "interaction model" on
>> > creation of the resource using
>> > Link:<http://www.w3.org/ns/ldp#Resource>;rel="type",
>> > that the created resource will ONLY have LDPR interaction model and not
>> > LDPC (ie containment and membership triples will not be affected by
>> POSTing
>> > to it or DELETE'ing any of the member resources) even though the entity
>> > body may have a triple where rdf:type of ldp:Container [1].
>>
>> I also have a small comment that I've been meaning to send regarding the
>> interaction model.
>>
>> 5.2.3.4 states that "Clients use the same syntax, that is HTTP Link
>> headers, to specify the desired interaction model when creating a
>> resource as servers use to advertise it on responses."
>>
>> I noticed that in the primer, the POST request to an LDP-BC does not
>> contain a link header expressing the type of the resource to be created.
>> That also seems to be the behaviour of test suite. However, the POST
>> request to an LDP-DC *contains* the Link header Link:
>> <http://www.w3.org/ns/ldp#Resource>; rel="type", while *none* of the
>> examples in the LDP spec show a Link header being sent with POST requests.
>>
>> Is there any reason for this behaviour? I would expect that clients
>> would normally send a Link header with the type of resource to be
>> created with every POST request. The way it is now, a person
>> implementing the spec will assume that by default, a POST request will
>> only create LDPRs (maybe that was intended but it never got documented?).
>>
>
> After reading 5.2.3.4, I was also in the opinion that clients should send
> a Link header with the interaction model when creating resources because it
> says "This specification does not constrain the server's behavior in other
> cases." i.e. in cases where the interaction model is not specified.
>
> There was a comment to the editors of the primer and we discussed whether
> we need to send the interaction model header in cases where the server can
> guess it easily. For example, non-RDF resources in which only the LDPR
> interaction model is applicable. Except for the case where we want to treat
> an LDPC content with LDPR interaction model (for archiving etc) the most
> other cases, the server can correctly guess the interaction model looking
> at the content. However, I still think according to the spec it is better
> to include the interaction model header in creation requests according to
> the spec.
>
> We would like to know what others think so we can do the necessary changes
> before republishing the the Primer draft.
>

I think requiring the header on create was not intended and not desirable.
IF the header is present AND the server can honor the request, then the
client overrides whatever the server would have done based on the content.
So I think that it makes complete sense for LDP servers to determine the
interaction model based on the content of the creation request, with the
Link header being part of that.

I believe the non-normative sentence of "This specification does not
constrain the server's behavior in other cases." is too vague and
overreaching.  Obviously the specification does constrain the server's
behavior in other cases.  It would probably be better to say: "When the
request header Link: rel='type' is omitted, the server determines the
interaction model based on the content of the entity body."  Again a
non-normative statement.

Regards,
Steve


>
> Best Regards,
> Nandana
>