I agree filtering on a single entity is silly:
But under the table of section 7.9 of the spec at
http://opensocial-resources.googlecode.com/svn/spec/draft/REST-API.xml

"

/people/@me/@self? filter...@friends&filterOp=contains&filterValue=<someUserId>
                                                

This will return nothing if the other ID is not a friend, the current user if the two are friends. filterValue may take a specific person identifier of @owner or @viewer.

"

Which implies that filtering on the *single* @me/@self entity is the way to find if the current user is friends with the specified user.

Back in the early thread, (about a week ago) I think I said the spec looked odd.

This is where the original URL that started this thread comes from.
---------------------------

So a question for you, Chris,
should /people/@me/@self respond to filtering?



Ian



On 13 Feb 2009, at 12:58, Chris Chabot wrote:

There's really no need to assume anything, /people/@me/@self is a single
entity and can never be more then one result.

Hence filtering is a bit silly, what exactly would you want to filter on,
what would be the use-case for wanting to know a viewer *only if
<condition>* ?

The only situation in which /people/@me/@self returns an error is when your using 2 legged OAuth (or anonymous read access is allowed), in which case the securityToken's getViewer() will throw an exception since @me can't be
resolve to an ID in those situations.

In every other situation, getViewer() returns an id, which is used to
resolve @me/@self to a single user record, which is not really filterable,
since filtering only makes sense on collections, right?

On Fri, Feb 13, 2009 at 1:52 PM, Ian Boston <[email protected]> wrote:

Can I unwind this a little please.

The original patch for 904 was to introduce a change the SPI for getPerson
with filtering.
I argued that getPeople already provides the filtering so there was no need
for a SPI change.
Ben pointed out that the REST response would then be a collection which
exposed a tension in the spec.

Did I get that right?

Lets assume for a moment that the output of the current servlet *is*
correct (ie not a collection)

having called getPeople check the response and emit a 404 if there are 0 entries and the only entry if there is 1 (there should not be more than 1 as
a result of the filter). Obviously the Future needs to be wrapped to
evaluate this when its needed and not before.

Does that make sense?

I think if they current servlet is not correct then that needs sorting out separately, but I see emails on this thread stacking up faster than I can
type or think or answer the phone.

Ian


On 13 Feb 2009, at 12:36, Chris Chabot wrote:

On Fri, Feb 13, 2009 at 1:27 PM, Ben Smith <[email protected]>
wrote:

On 13 Feb 2009, at 12:09, Chris Chabot wrote:

'@self' is not a collection by any definition of the word, it's a single

object, hence the lack of the collection (and the lack of filtering,
there
is only one 'self' and that's the viewer :)

The same goes for activities, a query on /activities/@me/@self returns a collection of activities (since it can contain 0, 1 or many results),
however /activities/@me/@self/<activityId> can only return 0 or 1
results,
so it returns an activity record and not a RestfulCollection.


But that's not what the spec says should happen: If the request is
specifically for a single contact (e.g. because the request contains Additional Path Information like /@me/@all/{id} or /@me/@self), then
entry
MUST be an object containing the single item returned (i.e. "entry": [ {
/*
first item */ } ] and "entry": { /* only item */ } respectively).

Unless I'm reading it wrong, /@me/@self should still return
<response><entry><person></person></entry></response>




There are 2 parts to the definition:

1) if the request could possibly return multiple items (as is normally the
case), this value MUST always be an array of results, even if there
happens
to be 0 or 1 matching results

if something could return multiple entries it has to be a collection,
however /@me/@self is a single entry, so this doesn't apply here

2) If the request is specifically for a single contact (e.g. because the
request contains Additional Path Information like /@me/@all/{id} or
/@me/@self), then entry MUST be an object containing the single item
returned

So if the query is for /@me/@self it *MUST be a single entry.*

I think your either confused and consfusing /@me/@self with /@me/ @all
(which
is an PortableContacts way of saying /@me/@friends), or your a believer in possessions and believe that our definition of a person being a singular
entity is wrong and the spec should be changed to allow for multiple
spirits
and/or demons in one physical body. (And i truly hope that is not what
your
suggesting :)



I agree that filter...@friends&filterValue=<userId> is confusing, but
filterBy=id&filterValue=<userId> would surely remove all people that do
not
have <userId> as their userId! I guess you really want a filterBy AND
filterOn (or something), or explicitly name filterValue to
filterUserIdValue
(or something).


Filtering on /@me/@self is confusing, since it's only one record, what
would
you want to filter on?




Reply via email to