Re: httpRange-14 Change Proposal

Jeni,

First, thanks for confirming - many responses in line from here:

Jeni Tennison wrote:
 > The server *can* return the same content from the /uri URI and from
 > the /uri-documentation URI, but it does not have to, and it wouldn't
 > be sensible to do so for an image. Your first question asked if the
 > server could return the same content, your second asked if it must.

Apologies for any confusion from my wording, however I did mean "can" 
rather than "must".

In a nutshell then, this proposal says that you can return a 200 OK for 
a GET request on any URI, but if you return "a representation of a 
description of the thing referred to by <uri>" rather than "a 
representation of the thing referred to by <uri>" then you should say it 
is so by including the special "<uri> :describedby <uri-documentation>" 
triple.

Additionally, rather than special casing this so that this rule let's a 
publisher override the default 200 OK return a representation of a 
resource, the proposal also aims to change web arch and the HTTP 
specification such that a 200 OK in response to a GET no longer returns 
a representation of the requested URI, rather it just returns a 
representation which you must consult to find out what it is.

That's quite a large change to the web / web arch / http.

> On 28 Mar 2012, at 16:07, Nathan wrote:
>> Jeni Tennison wrote:
>>> Yes, that's correct. With no constraining Accept headers, it could alternatively return HTML with embedded RDFa with a <link rel="describedby"> element, for example.
>> Is that universally true?
>>
>> Suppose /uri identified a PDF formatted ebook, or a digital image of a monkey in JPEG format, or even an RDF document.
> 
> Then it would return those things. I think that you may have leapt to the conclusion that /uri *always* returns the same as /uri-documentation. There's nothing to my knowledge that says that, indeed given that you can have several :describedby links it would be impossible.

Sorry no, not *always* just *always could* or *always can*. As in, it 
would be universally true that for any successful GET request you would 
receive a representation, and that representation may be a 
representation of the <target-uri>, or it may be a representation of 
<some-other-uri> which describes the target-uri.

>> Question A:
>>
>> Currently we have:
>> <http://example.org/uri>; - a JPEG image of a monkey.
>>
>> When you issue a GET on that URI the server currently responds
>> 200 OK
>> Content-Type: image/jpeg
>> Link: <http://example.org/uri-documentation>;; rel="describedby"
>>
>> So under this new proposal, the server can return the contents of /uri-documentation with a status of 200 OK for a GET on /uri?
> 
> Under the proposal, the server would return the JPEG with a 200 OK for a GET on /uri. http://example.org/uri-documentation would return a description of the JPEG in some machine-readable format. 

Or more accurately, the server MAY return the JPEG with a 200 OK for a 
GET on /uri, or it may return the same result as a successful GET on 
/uri-documentation (a description of the /uri in some machine readable 
format).

Is this limited to machine readable format, why not human readable too?

It appears that if one can return text/turtle for a GET request on 
</foo>, where { </foo> a :Horse } then one should also be able to return 
an image/jpeg which visually describes the horse.

>> If yes, this seems like massively unexpected functionality, like a proposal to treat "Accept: some/meta-data" like a DESCRIBE verb, and seems to exaggerate the URI substitution problem (as in /uri would be taking as naming the representation of /uri-documentation).
>>
>> If no, where's the language which precludes this? (and how would that language go, given that it's exactly the same protocol flow and nothing has changed - other than the reader presuming that /uri now identifies something that does have a representation that can be transferred over HTTP vs identifying something that doesn't have a representation that can be transferred over HTTP).
> 
> I don't really understand what you think it needs to say I'm afraid.



>> Question B:
>>
>> How would conneg work, and what would the presence of a Content-Location response header mean? Would HTTPBis need to be updated?
> 
> I can't see any way in which any of that would work differently from currently.

Okay, given the use-case of a GET on </uri> returning 200 OK, and the 
response containing a representation of </uri-documentation> in text/turtle:

What would the value of the Content-Location header be? /uri-documentation?

short version: this proposal would mean many sections of httpbis would 
need to be reworded and changed, as it conflicts to the point of saying 
the opposite.

>> Question C:
>>
>> Currently 303 "indicates that the requested resource does not have a representation of its own that can be transferred by the server over HTTP", and the Link header makes it clear that you are dealing with two different things (/uri and /uri-documentation), but where does this proposal make it clear at transfer protocol level that the representation included in the http response is a representation of another resource which describes the requested resource (rather than it being as the spec defines "a representation of the target resource")?
> 
> The proposal says that applications can draw no conclusions from information at the transfer protocol level about /uri. In particular, it can't tell whether the representation that is returned with /uri is *the content* of /uri or *the description* of /uri. Further information about /uri (eg that it is a foaf:Person) may help the application work out that the representation was *a description*.

Wow, so every URI no longer refers to anything unless it's explicitly 
stated in some RDF somewhere, and if one looks up <b> in a browser and 
sees a picture of a monkey, they are incorrect for saying it refers to a 
picture of a monkey if some RDF document somewhere describes <b> as a 
:SpaceShuttle.

Can the TAG really just say "okay, all http:// URIs no longer refer to 
anything"?

> However, an application can draw conclusions about /uri-documentation, assuming it gives a 2XX response, because it has been retrieved as the result of following a :describedby link (or if it were the target of a 303 redirection). The application can tell that the representation from /uri-documentation is *the content* of /uri-documentation and *the description* of /uri.

I can't see how it could tell that "the representation from 
/uri-documentation is *the content* of /uri-documentation and *the 
description* of /uri". Perhaps that it's *a* description of /uri, but 
certainly not that it's "the content of /uri-documentation", the 
proposal itself removes all notion of a representation being a 
representation of the current state of the requested uri.

if <a> is described by <b>, and <b> is described by <c>, then a GET on 
<a> can now return <b>, whilst a get on <b> can return <c>, and so 
forth, and if that :describedby triple is missing, or you don't get back 
RDF in some form, then you don't know what you retrieved or if the 
requested uri refers to it at all.

>>> Either way, there is no implication that what you've got from http://example.org/uri is the content of http://example.org/uri (or that http://example.org/uri identifies an information resource), but there is an implication that what you get from http://example.org/uri-documentation is the content of http://example.org/uri-documentation (and that http://example.org/uri-documentation is an information resource).
>> Sorry I don't follow, how is there an implication from a 200 OK for <uri-a> that it's not an IR and for <uri-b> that it is an IR?
> 
> Because /uri-documentation was reached through a :describedby link. This extra information allows the application to draw the conclusion that the representation from /uri-documentation is *the content* of /uri-documentation.

and when you don't reach it via a ":describedby" link (as in 99.99% of 
cases on the web)? also see above, same points.

>> If there was a Set of all Things (Set-A), then that set would have two sets, "the set of all things which can be transferred via a transfer protocol like HTTP" (Set-B), and then everything else (Set-C) which comprises Set-A minus Set-B. As far as I can tell, the one thing that determines whether something is a member of the Set-B or Set-C, for HTTP, is that 200 OK in response to a GET, hence why we need the 303.
>>
>> This proposal appears to try and override that "rule" (fact) by saying let the content of a representation define what is a member of Set-B or Set-C, however the act of dereferencing itself is what determines whether an identified thing is a member of Set-B, as Set-B is the set of all things that can be dereferenced. Hence my confusion at this proposal.
> 
> The "fact" that a 200 OK determines whether something is a member of Set-A or Set-B is a design choice made by httpRange-14, not a fundamental truth of the universe. The proposal makes a different design choice, in saying that you need more than just a 200 OK response to say, beyond all doubt, that a URI refers to something that is member of Set-B.

Apologies but I have to disagree completely here, I can say I'm a 
goldfish but I have the properties of a human and belong in the Set of 
Humans, no matter how much I say, I'm never going to be a goldfish - 
there's no design choice there, similarly if something a representation 
of something was retrieved via HTTP, then it belongs to the set of 
things which can have their representations retrieved via HTTP, that 
just is a fact, not a design decision.

Sorry this appears so negative, but... well the above hopefully 
explains, personally I see it as ripping the foundational constraints of 
the web/uris/http away to try and save an extra GET request in a few cases.

Nathan

Received on Wednesday, 28 March 2012 21:23:53 UTC