Hi Paul, I've included some responses inlined below.
> From: Paul Komar/Lexington/IBM@IBMUS > To: [email protected] > Date: 01/06/2011 04:16 PM > Subject: [oslc-core] What can an OSLC spec claim? What can a client assume? > Sent by: [email protected] > > Last month Martin Nally wrote " > I think this confirms that the only safe option for a client is to assume > nothing. I think the spec should say this." > > Can you please help me understand how a person can implement a client of a > service whose specification is based upon that much ambiguity? > (I got the impression from reading some historical notes from WebDAV client > implementers that while the spec might be written to allow varying degrees of > sophistication of service implementations, they would have preferred a > tighter, more constrained spec.) The intent of this guidance is to make it very clear that the motivation is to build flexible integrations that span time, upgrades, technology choices, etc. So for some scenarios, clients may depend on certain resource formats and expect certain properties. That won't change. OSLC style of integration again focuses on loosely-coupled integrations, whereas efforts like WebDAV focused on very specific client-server interaction of SCM tools. > Arthur Ryman replied to "I agree that clients should be able to gracefully > handle unexpected > responses." > It seems to me that clients can gracefully decline to provide the desired > behavior when the server doesn't give the client what it needs. > > Consider a use case where a client must get information from a service and > then use the information in the response to get more information (possibly > from another service). If the first service does not provide the requested > information, then the client cannot complete the use case. > > Also, I'm curious to learn how to write good compliance tests if the client > can assume nothing. In having just contributed a test suite, the test suite can assume it is speaking to services that claim a level of support. A test suite is not necessarily a well-behaved client in this way. > While the Rest/OSLC world may want to be more flexible than the UML world > (with its effective requirement of transitive closure across domains, as > Martin also wrote about), OSLC services must provide the information in a > consumable way, right? I thought that the Restful way to handle version > upgrades was to use media types that provided compatible ways to add > information in subsequent versions, but allow older/smaller clients to use the > older/smaller content. > > It seems to me that a Rest/OSLC client might have to have a "broader" > expectation of valid responses than an client from the UML world. (Perhaps > there's an equivalence class of useful responses?) > > In summary, I wonder if Martin or Arthur could write up a pattern for good > OSLC client implementations. > Maybe there would be a rule like " don't validate that the response is what > you expect, instead look in the response for the information that you require"? As you stated it is not a bad way to think about it. The Core WG is starting to build up primers and other material to help guide implementers. This is good input to that effort for consideration on writing well-behaved OSLC clients. Thanks, Steve Speicher | IBM Rational Software | (919) 254-0645
