[
Lists Home |
Date Index |
Thread Index
]
Somebody just shoot me, I'm getting into another email debate about
REST :-) That's one New Years Resolution down the drain....
> The real problem is that cookies are completely contrary to the web
> architecture. They create resources that do not have URIs, and thus
> cannot be bookmarked, linked to, and otherwise referenced. They
> attempt to force state into a fundamentally stateless protocol with
> disastrous results. Even if there were no privacy implications,
> cookies would still be the wrong solution... But there is no reason I
> should simultaneously give up the ability to bookmark or link to a
> page as a result of providing my name.
>
How can something as widely used as cookies be "contrary to the Web
architecture?"
One can say that they're contrary to the *theoretical* principles of
REST, but then the Web As It Is would be largely non-RESTful, eh? The
string "cookie" does not appear in the TAG Webarch document, so I don't
think they have weighed in on the subject.
One might argue that that Web would be better if it were more RESTful,
but that is an unproved assertion and not a logical deduction or
empirical generalization. One can of course argue on pragmatic grounds
that a specific site would be more useable if the results of a URI
dereference were always the same independently of the state maintained
on the client (e.g. a cookie), but compelling use cases might be hard
to find in a world where there tends to be 1 person for 1 computer (or
account on a computer).
For example, everyone who subscribes to Bloglines gets a different HTML
page when dereferencing http://www.bloglines.com/myblogs ; a cookie
somehow identifies *my* subscription list. (The state is maintained on
the server, so my various computers see the same state of the same
list). Is that a problem? I prefer it to having to dereference
http://www.bloglines.com/mchampion666 or whatever [I have a daughter
whose name begins with "M" and we are already competing for the
"mchampion" namespace!]. Contrary to the Web architecture? Arguably
it's just the "representation" of the resource *my* blogs that changes
with the definition of "me", just as (in the canonical example) the
representation of the weather page in Oaxaca changes with the weather.
I'm happy to argue that typical Web *services* should be more Web-like
(support GET, use hyperlinks, etc.), but I'm not so comfortable arguing
that typical Web *sites* should be come more RESTful, at least until
there is empirical evidence that RESTful web sites are in some tangible
sense better than non-RESTful sites.
|
