- From: Dave Kristol <dmk@allegra.att.com>
- Date: Wed, 16 Aug 95 15:57:49 EDT
- To: fielding@beach.w3.org
- Cc: http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com
Roy Fielding <fielding@beach.w3.org> wrote: > dmk@allegra.att.com (Dave Kristol) wrote: > >Section 8.7 of draft-ietf-http-v10-spec-01.ps says: > > > > If multiple encodings have been applied to a resource, the > > encoding-mechanisms must be listed in the order in which they > > were applied. > > ==== > > > >I think this is backward. I think they should be listed in the order > >in which they must be applied by the client in the order they're listed. > > ==== == > > > >Consider foo.gz.Z. Gzip was applied first, then compress. NCSA HTTP > >1.3 produces this header: > > Content-Encoding: x-compress, x-gzip > >(It looks like the old CERN code I have doesn't support multiple > >encodings.) > > I didn't think anyone did, but that interpretation is ugly. Ugly or not, it exists. Whether anyone actually uses it is another matter. > Message header field values should be "added" via an append > operation. Thus, each encoding becomes layered, and > > Content-Encoding: first > Content-Encoding: second > > has the same semantics as > > Content-Encoding: first, second > > If we reverse the order, then added headers must be prepended. That makes sense (aesthetically) if you assume that these encodings are applied dynamically. Often they have already been applied to static files (e.g., my example), in which case you have to reconstruct the encodings outside-in. From an implementation standpoint, it's easier to do it the way NCSA has done it. Dave Kristol
Received on Wednesday, 16 August 1995 13:09:28 UTC