- From: Chris Knight <Christopher.D.Knight@nasa.gov>
- Date: Thu, 10 Jul 2003 10:18:58 -0700
- To: Geoffrey M Clemm <geoffrey.clemm@us.ibm.com>
- CC: "Nevermann, Dr., Peter" <Peter.Nevermann@softwareag.com>, "'w3c-dist-auth@w3.org'" <w3c-dist-auth@w3.org>, w3c-dist-auth-request@w3.org
Geoffrey M Clemm wrote: >Good question. I would argue that 1 copy is the more desireable >behavior, but 2 copies will be easier for many servers to implement >(for example, that is what "cp -r" does on Unix). > >So my preferences would be (from high to low): >1- require 1 copy >2- leave it up to the server >3- require 2 copies. > >Any votes? > I'd definitely go for 1 (maintaining bindings with a copy) but I understand the server implementation implications are a bit of a challenge, especially for fs-backed servers. I'd strongly discourage 2 (maybe make the behavior a SHOULD instead of a MUST.) Moreso, I'd vote that it should be encouraged to maintain consistent behavior from the server (for intra-server copies) and the client (for inter-server copies). Certainly 3 is easier for clients to perform *unless you have loops*. I'm currently researching some implementation implications (using an RDBMS-backed server) of looping binds and it's not perty. It should be possible for a client (using the DAV:resourceid property) to identify if it's encountered a loop...But it would have to do copies step-wise as a PROPFIND Depth: infinity would result in a 506 Loop Detected error and incomplete/no information. Actually, it'd have to do this anyways whether it recreated bindings or "split on copy". 8^O
Received on Thursday, 10 July 2003 13:18:20 UTC