W3C home > Mailing lists > Public > public-ldp-wg@w3.org > November 2014

Re: Loosely-coupled (modular) LDP, was Re: Recharter scope

From: Sandro Hawke <sandro@w3.org>
Date: Wed, 19 Nov 2014 08:30:46 -0500
Message-ID: <546C9B86.50808@w3.org>
To: Steve Speicher <sspeiche@gmail.com>
CC: "public-ldp-wg@w3.org" <public-ldp-wg@w3.org>
On 11/18/2014 08:59 AM, Steve Speicher wrote:
> On Tue, Nov 18, 2014 at 7:27 AM, Sandro Hawke <sandro@w3.org> wrote:
>> On 11/18/2014 06:36 AM, henry.story@bblfish.net wrote:
>>> During the teleconf there was a discussion about a twitter thread
>>> criticising the RESTfulness
>>> of LDP. I think that it is the most important value that LDP bring to the
>>> table, when
>>> compared to all the other RDF standards that exist. We should think of
>>> this group as tying
>>> loose ends together.  Could someone please point us to that discussion?
>>> Note that I don't
>>> think that RESTafarians have the final say  of what REST is, because they
>>> tend
>>> to miss the semantic part of what it is that a Representation transfers:
>>> the state of
>>> a Resource. So we should not allow us to be buillied by bad RESTafarian
>>> arguments into
>>> abandoning the core value that REST represents for the LDP group.
>>
>> There is this:
>> http://lists.w3.org/Archives/Public/public-ldp/2014Feb/0020.html
>>
>> in which Mark Baker say, "I consider the design overly complex, tightly
>> coupled (and therefore unRESTful)".
>>
>> I had a private conversation with him after that, trying to understand, and
>> came to the conclusion that he was right.  In the simplest terms, we
>> shouldn't have any notion of an "LDP Server" or "LDP Client".   In general,
>> we should just have certain kinds of resources that behave certain ways.
>> An LDP Server could be defined as a web server that happens to support one
>> or more of the ldp container types, or something, for market simplicity.
> The HTTP spec have a notion of "HTTP Client" and "HTTP Server". An LDP
> Server doesn't have to support the entire LDP spec, I see it as more
> of a convention as a way to refer to the piece of software that has
> been written to receive requests that match the expectations as
> outlined in the LDP spec.
>
>> For example, I think it's possible (and preferable) to express
>> ldp:BasicContainer without talking about server or client conformance, and
>> without any RDF dependency, in a way that would make them more generally
>> useful.
> I'm not sure what good a standard is without conformance, I believe
> I'm missing the point somehow.

Yeah, I didn't explain this very well.  Let me try again.

I think it's analogous to code refactoring to improve modularity and reuse.

What we have now is specified as if it were a monolithic function with 
many toggles for optional features.  A "loosely-coupled" alternative 
would look more like a bunch of independent functions, each useful in 
its own right, which happen to work nicely together.

Mostly this is just about presentation in the spec, but in some cases my 
sense of a good refactoring would also change interoperability, eg 
splitting Container into Creator + PageSet.  In a sense this is a total 
coin-flip decision: do we want everything to be a container, where some 
containers create but don't act as sets and some act as sets but don't 
allow creation, or do we want two different interfaces that might be 
implemented by the same resource?     Right now that's kind of an 
annoying "who cares" question, but as each of those functions grows in 
complexity, with different ways one might want to use it, the overall 
system probably stays simpler if Creators are defined in one spec and 
PageSets are defined in another.    At least, that's the loose-coupling 
hypothesis, as I understand it.   (This might not be the best example: 
separating patching from containers is a much more appealing example, 
and we're all convinced that's a good idea, I think.)

On conformance/servers, it's probably just editorial, but it has a 
somewhat different feel.  Something like:

1.  When an LDP Server responds to a GET on a Creator, it MUST include a 
Link <...Creator> rel=type" header.   In handling a POST, the server 
MUST make the posted content available at a new URL.

vs

2.  A Creator resource is any resource whose server responds to POST 
requests on that resource by making the posted content available at a 
new URL.   In responding to GET and HEAD requests on a Creator, its 
server MAY include a Link <...Creator> rel=type" header to signal to 
clients how POST will be handled.  HTTP Servers MUST NOT include such a 
Link header on non-Creator resources.    A conforming "LDP Server" is 
one which always includes all the optional LDP headers.

In a sense the first only applies to LDP servers and the second applies 
to all HTTP servers, but I think there's an unwritten (?) rule, "HTTP 
Servers MUST NOT include Link headers unless they understand them and 
agree with what they mean", which makes the practical meaning of (1) and 
(2) the same.

>> Actually, even better, I'd probably go with:
>>
>> ldp:Creator -- a resource which responds to POST by making the POSTed entity
>> available as a new Web Resource.   Relative URLs inside the entity or in
>> Link headers are understood to be relative to the new location.
>> ldp:Creators may have an associated ldp:Container in which created resources
>> automatically appear (or may, in fact, also be a an ldp:Container).
>>
>> ldp:Container (or ldp:PageSet or ldp:Enumeration or something) -- a resource
>> which is is said to "contain" other resources.  Its state includes these
>> containment relations, so a GET returns an enumeration of its contained
>> resources, possibly along with other data, in an appropriate media type.  In
>> RDF serializations, the container ldp:contains each of its contained
>> resources.  Variations on (subclasses of) ldp:Container control what happens
>> when the container or a contained resources is DELETEd, whether
>> subcontainers are allowed, whether resource in a subcontainer are considered
>> elements of the parent container, etc.
> Personally I'm not sure this changes much from where we are at, just a
> different way to express it but not seeing how this would change
> "market acceptance" or something the rechartered group would need to
> do.   I think reflection is healthy to evaluate how to improve on the
> future.  Though some of this discussion seems less constructive on how
> we take the work we have agreed on thus far and work together to get
> support/adoption (market adoption) and feedback.

I haven't had a lot of conversations with people who had heard of LDP 
and were not in this WG, but in the ones I have had, they seemed to 
think LDP was really complicated.   It's the problem of people thinking 
they need to understand the whole menu before they can order a dish.    
The modular approach would make it more clear, I think, that they only 
need to read the specs for the bits they want.   This will become much 
more important as we grow the functionality.   Of course we're already 
doing this with Paging and Patch.

>> These are much more general concepts, not really even tied to each other,
>> not tied to RDF, etc.  They are so general they are already in widespread
>> use -- they're just not named, so clients can't detect them to allow for
>> interoperability.
> I don't understand this last point, the concepts or so general or the
> deployment of a spec supporting the concepts are in widespread use?
> Also I don't get the point about clients can't detect them, to allow
> for interoperability.  I would have expected you to say, that clients
> that don't know about them can continue to work (graceful
> degradation).

I hope the way I phrased defn (2) above makes clear what I meant. 
Creators, PageSets, Pagables, etc already exist on the Web, but they're 
not telling clients (in machine readable ways) that that's what they 
are.  (And they differ in arbitrary ways.)

To phrase this one more way: loose coupling is more like paving the 
cowpaths, or letting people pave their own cowpaths.  LDP 1.0 is more 
like a flexible order form for paved paths.   For a lot of people the 
order form is nicer, but it might turn out to be harder to maintain, and 
it gives people less flexibility for mixing-and-matching with other 
products.   That mixing-and-matching is a hallmark of the Internet and 
the Web, and that's the banner I think the RESTafarians are flying.

In terms of actually moving forward?   I think we should expect the next 
WG to produce 5-15 somewhat independent Recs.   Maybe after we've gotten 
the hang of that, and like that style, we can go back and re-articulate 
LDP 1.0 as several smaller specs.

This could also shape the way we talk about the new group and the 
existing work to the world, although nothing specific springs to mind.

       -- Sandro

>
> - Steve
>
>>        -- Sandro
>>
>>>> On 17 Nov 2014, at 21:24, Arnaud Le Hors <lehors@us.ibm.com> wrote:
>>>>
>>>> Per our discussion on today's call I invite everyone to help develop the
>>>> list of questions that defines the scope of the new WG we would propose to
>>>> charter. We could change the format if we want to but that's secondary. Here
>>>> is the updated list based on today's discussion:
>>>>
>>>> 1.        How can retrieval of a container and its contained resources be
>>>> combined so that fewer HTTP operations are required than it is necessary
>>>> with LDP 1.0?
>>> I think this needs to be formulated more clearly.
>>>
>>> As I understand it the aim would be to pass some information about the
>>> contents of the ldp:contain-ed resources
>>> in the LDPContainer. In some way Atom does this using the <content> and
>>> <summary> elements. What is desired
>>> is that this be part of a GET operation ( and not part of a query )
>>>
>>> => 1. define a way to allow  a container to respond in GET requests with
>>> full or partial content of the contained
>>>      resources.
>>>
>>>    ( clearly this either requirs literals or quads )
>>>
>>>
>>>> 2.        How can multiple resources be created with a single HTTP
>>>> request?
>>> Robert Sanderson wrote:
>>> "+0.  This seems like going firmly against the current specification that
>>> the entity body is a single resource, even if it refers to other
>>> non-existing resources on the same server, either by absolute or relative
>>> URIs."
>>>
>>> The really simple way to do this would be to POST a tar of a directory to
>>> an LDPContainer.
>>> What the working group should consider immediately is: does this
>>> contradict our current containers?
>>> Ideally POSTing a tar should create a LDPC with each of the tared files in
>>> the created LDPC.
>>>
>>>> 3.        How can a client request filtering what part of a resource or
>>>> container the server is to return?
>>> Robert Sanderson wrote:
>>> "+1.  And to keeping it separate from 1 and not using a query syntax"
>>>
>>> That inevitably will end up being some form of query syntax, if we have
>>> something
>>> general enough that it is worth specifying. What I think is desired is
>>> that there
>>> be a very simple query syntax available, that anyone can implement, that
>>> has very
>>> very low complexity properties, but that is extensible over time.  Most
>>> importantly
>>> it has to be semantically explicit, so that a client can know what types
>>> of queries
>>> he is asking as he jumps across servers by following links. ( That is
>>> property/value
>>> queries such as those found in current html forms are not good enough ).
>>>
>>> (btw, this is another reason for QUERY/SEARCH methods, as that builds in
>>> the variablitiy
>>> of syntaxes )
>>>
>>>> 4.        How can a client be notified when something, such as a resource
>>>> or set of resources, changes?
>>> Robert Sanderson wrote:
>>> "+1. And to working with SocialWeb to see if ActivityStreams will solve or
>>> already solves the problem for us."
>>>
>>> That requires polling. We should definitively see how a polling solution
>>> can help.
>>> But sometimes active notifications are much more desireable.
>>>
>>>> 5.        How can a client find out whether a service endpoint, such as a
>>>> SPARQL endpoint, is associated with a resource or set of resources?
>>>> 6.        How can access to a resource be controlled?
>>>> 7.        How can a client have greater control of how paging is done
>>>> (size, sorting, etc.)?"
>>>> 8.        How can a client learn what property constraints there are when
>>>> creating or updating a resource?"
>>> +1 to all the above
>>>
>>>> 9.        How can we do efficient transfer of LDP resources, either some
>>>> initial set or rolling updates (feed) of changes?
>>> One wonders wether a feed of PATCHes that would allow someone to
>>> reconstitute the changes would be useful.
>>> This is what people are calling Event Sourcing
>>>
>>> ( just found this reference )
>>>     http://ookami86.github.io/event-sourcing-in-practice/
>>>
>>>> Please, propose changes and additions as you see fit, as well as
>>>> deletions if you think that's appropriate.
>>>> Thanks.
>>>> --
>>>> Arnaud  Le Hors - Senior Technical Staff Member, Open Web Standards - IBM
>>>> Software Group
>>> Social Web Architect
>>> http://bblfish.net/
>>>
>>>
>>>
>>
Received on Wednesday, 19 November 2014 13:30:55 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 6 January 2015 21:11:59 UTC