- From: Geoffrey M. Clemm <gclemm@tantalum.atria.com>
- Date: Sat, 27 Feb 1999 09:20:48 -0500
- To: yarong@microsoft.com
- Cc: w3c-dist-auth@w3.org
From: Yaron Goland <yarong@microsoft.com> What really scares me is a scenario where I have a directory filled with references but I'm using an RFC 2518 client. If I copy the directory I will go from a directory that took up a few kilobytes (just to record the references) to one of any random and potentially huge size. The source directory ate 30Kb and the destination eats up 6 Gig. I would call that surprising. What really scares me is when I create a copy, and it turns out that it's not really a copy, so every change I make to the copy trashes my original. Apparently we're scared by different things. I would also invoke precedent here. In every system I have ever heard of that supports references (read: links) a COPY always copies the link not the destination. I would be very hesitant to go against three decades of accumulated experience without a good reason. So you've never even *heard* of Unix? If you're looking for decades of experience with links, you might want to consider it. Every Unix system I've ever used (and I've used *a lot*), copy by default the target of both hard and soft links, not the link itself. This is try for both "cp" and "tar". So I agree with your appeal to precedent, but I suggest you drew the wrong conclusion. Among the various reasons why the link target is copied by default, is that any relative links outside of the collection you are copying will break if you just copy the link itself. Converting the link to an absolute link to work around this would be even worse. Hence I believe that the default action should be no-passthrough on COPY. I could live with either one, but I'd be careful about going against three decades of precedent that says otherwise (:-). Cheers, Geoff
Received on Saturday, 27 February 1999 09:20:52 UTC