- From: Henrik Frystyk Nielsen <frystyk@microsoft.com>
- Date: Thu, 25 Jan 2001 14:09:03 -0800
- To: "Mark Baker - Ottawa Consumer and Embedded Div." <Mark.A.Baker@canada.sun.com>, <xml-dist-app@w3.org>
- Cc: "Frank DeRose" <frankd@tibco.com>, <rayw@netscape.com>
> >What would be just as reasonable would be to have the scenario > >play out this way: > > > > I send you an insurance claim with references to two photos of > > my broken car. I include the references to the photos using > > something like "http://my.com/brokencar..." URIs. > > > >Of all things that HTTP is being used for this is actually a > >good one :) HTTP is really good at GET! > > You're preaching to the choir! 8-) But I don't see why using MIME > necessarily means using a mid: schemes to identify parts. Why not use > the true URLs and Content-location rather than locally-scoped names? You have to be able to bind the two together somehow - otherwise you don't know that they are related. MIME doesn't have the concept of links other than the "Reference" header field which can only point to message id's - it is not a general linking capability so assuming that the links have to be listed in XML somehow, we can make the bindings like follows: XML document --> message id --> content-location URI XML document --> content-location URI --> message id In the first mapping the XML contains a mid: URI and the MIME entity with that message id has a content-location header field. In the second, the content-location URI is used directly in the XML *and* in the MIME entity. Both have the problem of establishing trust that the MIME entity really *is* an entity generated by the URI of the content-location but in addition the latter has the problem that the receiver has to parse every MIME entity to see if there is a Content-Location match for that URI. Especially when using nested MIME messages, this search can be quite elaborate. Say for example that in the car insurance example that photo B was of your car but that I sent a version of it that showed no marks "proving" that even if I hit you, your car didn't get a dent. One can say that this is the same problem that one has with HTTP caches - why should I trust that a cached response really comes from the origin server? The difference is that I as a client *choose* to trust the cache - I don't have to trust an arbitrary origin server that it does the right thing. Thanks! Henrik
Received on Thursday, 25 January 2001 17:09:35 UTC