[uf-rest] REST opacity and URL schemes.

[Breton Slivka]

It would seem that my discovery of a rarely used aspect of the HTTP  
url scheme, namely the semicolon ; has led to perhaps a level of  
unjustified excitement and advocacy. Intent is very difficult to  
gleam from carefully reading a spec, despite, or perhaps because of  
the high standards specs are held to for preciseness and lack of  
ambiguity, which encourage a sort of unnatural writing style which in  
all honesty is hard to read and easy to misinterpret. That aside, I  
have a question after reading a set of documents by Tim Berners-Lee,  
which are more geared toward the intentional side of things rather  
than the specification side of things, which you will find here: <  
http://www.w3.org/DesignIssues/Axioms.html >

Regarding the issues I'm concerned with, this is my interpretation of  
the document, and you can tell me whether it's incorrect or not:

*For REST, it is mostly important that GET have no side effects, and  
anything which does have side effects be implemented with POST (or  
mail).

*forward slashes, as mapped to unix heirarchical structures are used  
merely as a matter of convenience, and due to the principals of  
opacity, are not neccesarily significant to the resolution of a URI,   
except regarding the resolution of relative URI's in which case, such  
interpretation should be defined per URI scheme, with client support.  
Forward slashes do not indicate a resource, but the entire URI  
indicates the resource, and it's entirely up to the server how to  
find that resource based on the URL. No procedure for internal URI  
resolution is explicitly defined.

*That http defines such a scheme for resolving relative URI's for /  
style urls, but not yet for semicolons as indicators of parameters  
(though TBL provides such a scheme as a possibility for implementation)

*Therefore, virtually any symbol, including but not limited to  
forward slashes and semicolons can be used for the purposes of  
mapping to arbitrary data topologies, at the discretion of the URI  
scheme designer, which can be interpreted and resolved by the server  
in whatever way is appropriate. In addition, a great deal of this  
flexibility is left in the http scheme to be used at the discretion  
of server programmers. As long as opacity is observed, and URI's  
remain idempotent nouns.

*To this add the additional constraint given by roy fielding that  
this can only be done as long as the nature of the URI scheme is  
clearly communicated by the server to the client. In other words,  
there is no assumption made that the client will be able to infer a  
URI scheme from a base URI without explicit communication that such a  
URI scheme is in use.

*Notable exception is the # fragment identifier which is reserved for  
use by clients, and to be defined when registering the mime type for  
a document.

*TBL then demonstrates the flexibility of URI's by laying out his  
Matrix URI scheme, that is using the ; and = symbols to specify a  
matrix information space, which is distinctly incompatible with a  
heirarchical space. This does not imply that ; has any inherent  
meaning relative to the REST or uri model, but as a scheme designer,  
TBL is using it to represent a type of information topology being  
made available by a server.

If one is to take all the above statements as true and accurate  
interpretations of REST, then solutions to certain problems become  
available, which are not possible, or are awkward under a naive  
surface impression that REST requires URLS to contain only forward  
slashes / to represent information topologies. It is beginning to  
look to me that this impression has more to do with Aesthetics, and  
search engine technologies, than any specific requirements of REST.

A problem this interpretation potentially solves, for instance, is  
that of multiple views of the same information resource. Usually,  
this would be accomplished through content negotiation. However this  
is only effective if your multiple views all have distinct mime  
types. If you have for example, a resource such as < http:// 
example.com/blog/posts/53 > representing an article on a blog, you  
may have a non editable view, and an editable view.

In an ideal world this sort of thing could be specified using a  
fragment identifier,  such as < http://example.com/blog/posts/53#edit  
 >, however for this to work on a practical level, it requires client  
support which is not likely, and somewhat incompatible with the  
fragment identifier scheme defined for html/xml

Another option is to use the forward slash  < http://example.com/blog/ 
posts/53/edit >
However it seems logically innacurate to state that a view is a  
hierarchical child of the resource, rather than something more  
lateral, or directly related.

If the bullet points above are accurate, then it becomes possible and  
reasonable to use a semi colon, or any other appropriate symbol to  
represent a lateral space, such as a view. Matrix notation seems  
appropriate, since such a view system can be seen as a "matrix" (or  
tensor) with 1 axis, and english words representing named points  
along that axis.
< http://example.com/blog/posts/53;view=edit > < http://example.com/ 
blog/posts/53;view=display> < http://example.com/blog/posts/ 
53;view=hatom>. This makes sense to me as viewing a slice or  
subsection of a single resource, which includes multiple possible views.

The caveats being how relative URL's are resolved by clients, whether  
the server supports this level of flexibility, and whether  
appropriate mechanisms are employed for informing the client of the  
URI scheme in use (such as hrefs along all points on a matrix  
allowing navigation)

This may be overintellectualizing a simple problem, but the main  
purpose of this post is to verify if I have my bulleted  
interpretations correct, otherwise my conclusion may be false.