On 2006/06/29, at 12:24 PM, Mark Baker wrote:

Right now, sites can trust that referers from browsers, at least,
can't be manipulated. By changing that, you're reducing the
information about the request context available to them.

Right, but I don't think that matters much in the single domain case
since the author controls much of the request context.

Let's say I publish some content that I only want consumed by XHR that's sourced from my site. Sure, somebody can fake that request if they have control of the client, but I'm not looking for that level of control.

If I can't trust XHR to send a referer, I have to allow all requests, and that means that -- today -- somebody can link to that content from another site using <a>, <script>, <object>, etc.

Again, it isn't perfect; it's just a useful mechanism that's sometimes good enough (or at least better than nothing).

Turn it around: What's the use case for changing it? Why is it so
hard to send it?

AIUI, current practice is not to send the Referer header.

AFAIK the only browser that doesn't is Mozilla, and they're on track to adding it soon.
  https://bugzilla.mozilla.org/show_bug.cgi?id=170477

It's not hard to send, of course.  It's just a long string (commonly)
that doesn't add much value in the single domain case, IMO, but would
increase latency in the general case.  Not a good tradeoff, IMO.

Maybe we should get rid of that pesky Request-URI too, and move to a fixed-length object identifier... ;)

--
Mark Nottingham
[EMAIL PROTECTED]




Reply via email to