- From: Yaron Goland <yarong@microsoft.com>
- Date: Wed, 4 Sep 1996 11:37:03 -0700
- To: "'w3c-dist-auth@w3.org'" <w3c-dist-auth@w3.org>, "'www-vers-wg@ics.UCI.EDU'" <www-vers-wg@ics.UCI.EDU>, "'David G. Durand'" <dgd@cs.bu.edu>
Under the heading of "ME TOO" the requirements sheet I posted had Check Outs and Locks clearly separated. They are two totally different operations. Yaron >---------- >From: David G. Durand[SMTP:dgd@cs.bu.edu] >Sent: Tuesday, September 03, 1996 9:07 PM >To: w3c-dist-auth@w3.org; www-vers-wg@ics.UCI.EDU >Subject: Re: Seiwald Q & A -- "GET for EDIT" cookies > >At 4:38 PM 9/3/96, Christopher Seiwald wrote: >>| From: "Roy T. Fielding" <fielding@liege.ICS.UCI.EDU> >>| Checkout does not have meaning on all systems, whereas version has a >>generic >>| meaning (at times, too generic). It is intended to be as flexible as >>possible. >>| The "Content-" prefix is a requirement of MIME for what HTTP calls >>| entity-header field names. >> >>Not to beat a dead horse (but this one's still kicking): >> >>I'm pushing for us to recognise "checkout" as a meaningful act for the >>version abstraction we are trying to support, whether or not the underlying >>system can make real use of it. It means "GET the SOURCE for EDIT". >>"GET" because the document has to be fetched (or via existing methods >>determined that the user has the requested copy), "the SOURCE" because >>the user needs the document as it is prior to any server-side >>interpretation, and "for EDIT" so that the VC backend is aware of the >>user's activity. > >If we have a CHECKOUT method, then we don't need the LOCK method I propose. >But we must tell clients to ask for a checkout before trying a put, in case >they need one. We cannot require that clients do a special GET operation >before posting an update because it's not always required, and could just >send a lot of redundant bytes. A system is free to implement the protocol >so that sending the redundant bytes is a requirement, but I don't think the >protocol should require it. > > I myself don't see, nor have I heard any argument showing how my >proposal for a separate operation (wh/ probably should not be called LOCK) >to reserve a resource, separate from GETing it, is functionally inferior to >a joined-at-the-hip checkout that is not as flexible. Maybe the REQUEST(old >LOCK) operation needs a "GET required" status code for systems that want to >make me consume some fresh bytes. > > This brings me to a question. One of the points that I am most attached >to is the "configuration management treated separately" requirement. The >simplest foundation for any versioning system is to turn resource addresses >into ordered pairs of (ID x version). Once we have that we can implement >lots of policies on top -- the number of CM systems implemented on top of >RCS tends to support that claim. So I'd like to hold off discussions of >these complex policy issues until we have to get to them. And I think that >if Content-version can serve as a cookie, then it should, because it makes >the model for the simple stuff simpler, and doesn't add much work for a >complex system anyway. > >I'm afraid that with tabs changed to spaces by some mailer, your table of >policies was too hard for me to decipher. But I think that this needs to go >on hold. > >>There is use for the checkout cookie for all these systems, even if the >>cookied degrades into being little more than "Content-Version". Since >>All VC systems can make use of the cookies, and some need them for sane >>operation (checkin without checkout under clearcase is a no-go), it makes >>sense to use a single tag across all underlying VC systems. > >I think we want a wide variety of version styles to work nicely. I also >think that client requirements have to be simple to be widely implemented >-- and if they're not to be widely implemented outside the hard-core >configuration management community, then standardization is a waste of >time. Why is it bad for a "checkout-style" cookie to be the same thing as >the version-ID (ie. the Content-version header)? It works for complex >systems and simple ones, and is less work for simple systems. > >I think we may have actually run this to ground. Either we're talking about >a single field associated with a resource, and we're arguing about whether >to call it a Cookie or a Content-version, or we're claiming that two fields >are required. I don't see that we need two fields. In fact, I think that >the negotiation flexibility gained in the separation of resource >reservation from data movement makes it easier to and negotiate the >assignment of different version numbers at reservation and release times. I >even think treating a cookie as a "temporary version number for the working >version" is a nice concrete way to think of the semantics of parallel >sessions. If we really need two fields, then we can probably postpone the >discussion. If we don't, then I'd argue that Content-version is the name >that should win because it best represents the simple case. But I must say >that if we're just arguing about the name of a field I don't care, except >for the confusion that it will cause. > > -- David > >--------------------------------------------+-------------------------- >David Durand dgd@cs.bu.edu | david@dynamicDiagrams.com >Boston University Computer Science | Dynamic Diagrams >http://www.cs.bu.edu/students/grads/dgd/ | http://dynamicDiagrams.com/ > > >
Received on Thursday, 5 September 1996 13:11:37 UTC