W3C home > Mailing lists > Public > www-tag@w3.org > July 2009

Re: Review of new HTTPbis text for 303 See Other

From: Richard Cyganiak <richard@cyganiak.de>
Date: Mon, 13 Jul 2009 17:24:43 +0200
Cc: "Roy T. Fielding" <fielding@gbiv.com>, Jonathan Rees <jar@creativecommons.org>, Julian Reschke <julian.reschke@gmx.de>, HTTP Working Group <ietf-http-wg@w3.org>, www-tag@w3.org
Message-Id: <65857F7E-AC3F-4455-B98F-99C969016B70@cyganiak.de>
To: Pat Hayes <phayes@ihmc.us>
Pat,

1. You are on the web architecture mailing list. It should have  
occurred to you that I use the word "representation" in the web  
architecture sense: a stream of bytes with a content type (and maybe  
additional metadata). Your examples about your painted self-portrait  
and the number 17 miss the point. It's like insisting on discussing  
human anatomy in a thread about object-oriented programming languages  
because not all "members" are instance-scoped variables.

2. I said: "As far as HTTP is concerned, it does not matter much what  
the resource actually is -- a document, a file on a server, a person,  
whatever." You heard: "A URI can identify a person." What I meant was:  
"For HTTP, it REALLY DOES NOT MATTER what the resource actually is."  
It is simply the thing that is thought to sit behind the HTTP  
interface and that the HTTP conversation is about. You can implement  
the HTTP protocol, both on the server and client side, while being  
totally agnostic about what can and cannot be identified by HTTP. All  
you need to know is that URIs identify something that can have  
publisher-chosen representations. Wether that something actually is a  
document, file on a server, or a person, is IRRELEVANT to discussions  
of the HTTP protocol.

Because you misunderstood my point on these two counts, you end up  
ranting against a POV that I do not hold.

I especially continue to maintain that any talk about denotation is  
out of place on the HTTP protocol level. There is no such thing as  
denotation in the universe of the Hypertext Transfer Protocol. Yes,  
people obviously use HTTP URIs to denote all sorts of things, and a  
lot can be said about how one should model resources and  
representations based on the things one wants to denote, and what one  
can or cannot infer about the denotation of a URI based on HTTP  
interactions, but none of this matters one bit for the actual  
operations of the protocol. The protocol is just about pushing  
representations around.

Best,
Richard



On 13 Jul 2009, at 01:31, Pat Hayes wrote:
> On Jul 11, 2009, at 5:27 AM, Richard Cyganiak wrote:
>
>> Pat,
>>
>> On 10 Jul 2009, at 01:32, Pat Hayes wrote:
>>>> If the server has a transferable representation, it would
>>>> respond to the GET with the appropriate status code (200 or 304).
>>>
>>> Well, yes, IF it were driven solely by what one might call  
>>> rational HTTP architectural principles. BUt surely the whole issue  
>>> about httprange14 is that it introduces new principles which on  
>>> their face have nothing to do with http architecture as such, but  
>>> to do with denotation and naming.
>>
>> Not as far as HTTP is concerned. HTTP is just a transfer protocol.  
>> The HTTP world is really simple:
>>
>> 1. There are URIs. URIs are thought to identify things called  
>> resources.
>
> OK, stop there and tell me what you mean here by "identifies".  
> Because...
>
>> As far as HTTP is concerned, it does not matter much what the  
>> resource actually is -- a document, a file on a server, a person,  
>> whatever.
>
> ... in the usual sense of 'identifies' that one might expect to be  
> use in the context of a network transfer protocol, which is similar  
> to the use one might expect when talking about programming language  
> identifiers and what they identify, it most certainly does matter.  
> In particular, it simply does not make sense to speak, using that  
> normal terminology, of 'identifying' a person (or a galaxy or a  
> sodium atom, etc.); in fact, it does not make sense to talk of  
> identifying anything much beyond some kind of data structure or data  
> object. So if HTML claims to be able to make sense of talking of  
> 'identifying' people (for example), it must be in a wholly different  
> space than all previous computationally based notational systems,  
> and be using the word "identify" in a wholly different sense. And,  
> to repeat, can you tell me what that sense is?
>
>>
>> 2. Resources (whatever they are) are thought to have things called  
>> representations. As far as HTTP is concerned, it is totally up to  
>> the server owner to decide what's a representation of what. After  
>> the server owner has made their decision, a resource either has a  
>> representation or not.
>
> Really? OK, I will take you at your word. I am a server owner, and I  
> will decide that a certain resource, to wit, me, has a thing called  
> a representation of me. This representation of me is in fact a  
> portrait, painted using acrylic paints on a piece of masonite  
> approximately 30 cm square almost exactly a month ago: but let us  
> not go into details, as you tell me that such details are none of  
> HTTP's business. Still, the representation exists, and the resource  
> has it. OK, let us proceed.
>
>>
>> 3. If a resource has a representation, then a GET to its URI should  
>> be answered by 200. If not, then 303, 404 or 410 would be fine  
>> choices.
>
> So, HTTP must reply to a GET on my URI with a 200. OK, what should  
> it put as the payload of this 200 response, attached to the code  
> information? HOw do I get acrylic-coated masonite into an http  
> response? There is no representation which can be transmitted in  
> bits. You did not mention this aspect in your above summary: was  
> that an omission?
>
>>
>> I repeat: For the operation of the HTTP protocol, IT DOES NOT  
>> MATTER what exactly a resource is and what the exact relationship  
>> between resources and representations is.
>
> As you can see, I took advantage of this freedom in my example.
>
>> All these matters of denotation, information resources and so on  
>> are introduced by higher layers of the architecture.
>
> Wrong. Denotation is not introduced by a higher level, and even if  
> it were, it would not be higher in an architectural sense. You, in  
> this very message, in fact brought denotation into the picture, by  
> telling me that a URI can "identify" a person. URIs are symbols  
> strings, and the ONLY POSSIBLE SEMANTIC RELATIONSHIP between ANY  
> symbol and a physical object, is denotation. Sorry to shout there a  
> little, but the point needs to be made strongly. That is what  
> "denotation" means: it is all that is left of "identifying" when you  
> take away the actual network machinery, the computational byte- 
> transferring. And you have to take this away when you start claiming  
> to talk of relationships between names (of any kind) and non- 
> computational entities such as people (or indeed of any kind),  
> simply because computational byte-transfer talk is COMPLETELY  
> IRRELEVANT to semantic relationships (such as "identification")  
> between symbols (of any kind) and non-computational entities. The  
> fact, if it is a fact, that this word is not in your technical  
> vocabulary is entirely irrelevant. By claiming that your symbols  
> "identify" non-computational entities such as people or books (or  
> the weather in Oaxacala, to take another random example) , you are  
> no longer playing in the network-architectural sandbox, precisely  
> because these kind of things simply are not connected to networks in  
> the same functional sense that things like web servers are. Either  
> HTTP is a computational notion or it isn't. If it is, then it is  
> indeed quite simple. And I would be delighted if the HTTP literature  
> simply restricted itself to the computational world. But it does  
> not, and never has: HTTP has ALWAYS had these claims to semantic  
> grandeur: it has ALWAYS claimed to be not just about web sites and  
> web servers and files and documents, but about the whole grand span  
> of symbol usage to refer to absolutely anything in any possible  
> universe. And if indeed that is what HTTP is claiming to be able to  
> talk about, then it is about denotation, right out of the box.
>
>>
>> Yes, it would be useful to provide guidance to publishers about how  
>> best to model their information space as resources and  
>> representations. But this is out of scope for the HTTP protocol.
>
> See above. If indeed it is out of scope, so is any talk of URIs  
> "identifying" people. You can't have it both ways. Either you are  
> doing real semantics or you aren't. If you aren't, then don't make  
> ridiculous claims about "identifying" things that have no possible  
> connection to any physical network, or of "representations" that  
> cannot be sent in a byte stream.
>
>> The HTTP protocol kicks in AFTER the publisher has made up their  
>> mind about what resources they have and wether they have  
>> representations or not.
>
> OK, please tell me how to use HTTP to send my piece of masonite  
> attached to a 200 code. I've made up MY mind: over to you.
>
>>
>> Now, different subcommunities have different opinions on how to  
>> model resources and representations. That's not a good thing, and  
>> it would be good for interoperability if everyone agreed. However,  
>> this is pretty much orthogonal to any discussion of the HTTP  
>> protocol. As long as the subcommunities subscribe to the basic "URI- 
>> identifies-resource-which-can-have-representations" model, HTTP can  
>> accomodate them.
>>
>> Now let me take off my RDF hat for a bit.
>>
>> The suggested change for the 303 text came about because one  
>> subcommunity had the funny idea that some resources SHOULD have  
>> URIs but NO representations and it should STILL be possible to get  
>> information about them via HTTP.
>
> No, that is not the primary reason. Http-range-14 is not about  
> resources, it is about URIs and what they denote. The dilemma is  
> that people want 'normal' URIs to denote what it that HTTP thinks of  
> them as identifying, the "information resource" (not that that  
> matters). Which would be fine, except that there are some URIs which  
> people want to denote something else. And still, actually for  
> different ('linked data', Timblish) reasons, people want a GET on  
> those URIs to finish up, one way or another, with useful information  
> being returned. This is a problem. It would be ugly to have two  
> 'kinds' of URI, and impossible to change the millions of 'normal'  
> URIs in any way at all. The decision allows the few non-normal URIs  
> to take part in a slightly irrational HTTP dance which allows  
> everyone to say: look, since it didn't return a 200 code, its not  
> 'normal', and HTTP says it doesn't identify anything at all; so the  
> 'normal' assumptions about what it denotes are cancelled. And that  
> cancellation is the entire content of the decision: it has no other  
> purpose. The nature of the entity which handles the GET, and the  
> presence or absence of 'representations' of it, are irrelevant.
>
>> It beats me why anyone would want to do that
>
> The reason is that there are, believe it or not, entities in the  
> universe other than web servers; and people want to refer to them  
> using URIs.
>
>> ; but if we can make them happy with a minimal tweak to the  
>> language of an existing status code, then why not. HTTP is for  
>> everyone.
>>
>>> If the URI in the GET request is not intended to denote the  
>>> resource to which the GET is directed, then that resource must  
>>> issue a 303 redirection, and must not return a representation  
>>> using a 200 status code.
>>
>> There is no such thing as denotation in HTTP. The only relation  
>> between URIs and resources in HTTP is "identifies".
>
> Which, if i means anything at all when used between a symbol and a  
> non-computational entity, means 'denote' (or, if you prefer, 'refers  
> to' or 'is a name for'; they are all equivalent usages.) And again,  
> I challenge you (or anyone else) to tall me what "identifies" can  
> possibly mean, in thee circumstances, other than this.
>
>> If you care about other relations, you have to figure out how to  
>> translate them into the "URI-identifies-resource-which-can-have- 
>> representations" model of HTTP.
>
> That model is either (1) already about denotation, or (2) utterly  
> broken, or (3) meaningless as stated.
>
>>
>>> That has nothing to do with the existence or not of such a  
>>> representation. Even if the representation exists and the server  
>>> has access to it, it cannot return it with a 200 code when the URI  
>>> is intended to denote some other thing, in particular a non- 
>>> information resource of some kind.
>>
>> Wether a representation exists or not for a particular kind of  
>> resource is entirely up to the server owner, as far as HTTP is  
>> concerned. If you subscribe to a religion that says, "Thou shall  
>> not make a representation of me, for I am not an information  
>> resource", then that's great, and let me shake your hand brother,  
>> but this has no effect on HTTP.
>
> But thats the easy case. The hard case, for you, is when I use that  
> very handy English word "representation" is one of its normal  
> senses, not when I refuse to use it at all. There are many,  many  
> kinds of representations of things, and only a miniscule proportion  
> of them have anything even remotely to do with computers or network  
> transfer protocols.
>
>>
>>> If we follow your rule, above, and also httprange14, then a server  
>>> can be placed in an impossible position. If it has a  
>>> representation of itself which  could be put into a 200-code  
>>> response, and it receives a GET request with a URI which it knows  
>>> (somehow, perhaps by some externally agreed convention) is being  
>>> used to denote a non-information resource; what should it do?  
>>> HTTPrange14 requires it to not deliver a 200-coded reply, but your  
>>> criterion requires that it must. This is why I think the wording  
>>> should make absilutely minimal assumptions about what exactly the  
>>> 303 means.
>>
>> (RDF hat back on) Any sensible definition of "non-information  
>> resource" obviously MUST entail "does not have representations in  
>> the HTTP sense". In fact, that IS the definition of "non- 
>> information resource", in my book.
>>
>
> Of course, but that is completely irrelevant to my point. The  
> server, in my example, is not the non-information resource that the  
> URI refers to; that is precisely why httprange14 requires it, the  
> server, to emit a 303 code rather than a 200 code. It is merely the  
> servant whose job it is to emit the appropriate code to make  
> everything work properly. But it is AN information resource, and it  
> may well have a representation (in the http sense) of itself. Its  
> just a different resource than the one the URI denotes/refers to.
>
>> Wrapping up:
>>
>> For the function of the HTTP transfer protocol, it does not matter  
>> what exactly the nature of the things identified by URIs is.
>
> Oh, but it does. Because HTTP talks about information transfer  
> between entities which can transfer information, but it talks of  
> 'identification' of ANY THINGS WHATSOEVER, whether they can or even  
> possibly could transfer information. For example, a numeral  
> identifies a number, and also is a representation of it. So HTTP  
> should apply to this case as well, according to what you say here. I  
> should be able to send a GET request to the number seventeen and  
> expect to get sent back a 200-coded response with a suitable numeral  
> in its body, say "17". I know that is ridiculous: but it FOLLOWS  
> FROM WHAT YOU ARE SAYING; ergo, what you are saying is ridiculous.   
> So you ought to modify what you are saying, so that it makes more  
> sense.
>
>>
>> For the function of the HTTP transfer protocol, it does not matter  
>> wether the things you serve as representations on your server make  
>> particularly good representations of the resources.
>>
>> There are different schools of thought that try to clarify the  
>> nature of the "identifies" and "has representation" relationships,  
>> and this is critically important if we want to use HTTP URIs as  
>> identifiers for things that exist outside of the Web. But the HTTP  
>> protocol itself is and should be agnostic with regard to your  
>> position in these debates. That's layering.
>
> No, it is a poisonous combination of semantic (or maybe  
> philosophical or semiotic) ignorance, and hubris. You want http to  
> be universal, but you are claiming a kind of universality which goes  
> way beyond anything to do with network architecture, and so you  
> can't escape the consequences by appealing to network design  
> principles.  Maybe you don't intend to be doing this, but it is  
> being done by what you (and I should cast this in a kind of  
> anonymous plural, as the excellent southern phrase y'all, as I don't  
> intend this rant to be directed at you in particular) are actually  
> saying.
>
> Best wishes
>
> Pat
>
>>
>> Best,
>> Richard
>>
>>
>>
>>>
>>> Pat
>>>
>>>> ....Roy
>>>>
>>>>
>>>>
>>>
>>> ------------------------------------------------------------
>>> IHMC                                     (850)434 8903 or (650)494  
>>> 3973
>>> 40 South Alcaniz St.           (850)202 4416   office
>>> Pensacola                            (850)202 4440   fax
>>> FL 32502                              (850)291 0667   mobile
>>> phayesAT-SIGNihmc.us       http://www.ihmc.us/users/phayes
>>>
>>>
>>>
>>>
>>>
>>>
>>
>>
>>
>
> ------------------------------------------------------------
> IHMC                                     (850)434 8903 or (650)494  
> 3973
> 40 South Alcaniz St.           (850)202 4416   office
> Pensacola                            (850)202 4440   fax
> FL 32502                              (850)291 0667   mobile
> phayesAT-SIGNihmc.us       http://www.ihmc.us/users/phayes
>
>
>
>
>
Received on Monday, 13 July 2009 15:25:22 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Thursday, 26 April 2012 12:48:14 GMT