I've spent some time lately looking at the architecture for the next phase of implementing the Component Synchroniser for the Tridion Power Tools project. This meant looking through most of the other power tools, because, of course, they are a great resource for anyone wanting to build a Tridion GUI extension. The down side of this is sometimes, reading the code, you can observe a pattern being used, but it can be hard to tell why this would be a good or bad design. I'd noticed that the model of pretty much every power tool is implemented as a WCF service, often acting as a very thin wrapper around the core service client. As I was wondering about this, I posed the following question in the private chat channel used by the Tridion MVPs and community builders:

So if you're doing a gui extension, is it reckoned to be bad form to access the core service directly from your aspx. Or is it just coincidence that most (all?) of the power tools have an additional service layer?

This was enough to spark quite an informative debate, and in keeping with the spirit of the thing, I promised to write it up for general consumption. The contributors were Frank van Puffelen, Nuno Linhares, Peter Kjaer and Jeremy Grand-Scrutton.

The general feeling was that you ought to stick to the pattern I had observed in the power tools. The reasons were as follows:

• Ease of coding - The Anguilla framework can automatically generate a JavaScript proxy for your service.
• Maintainability - if you talk directly from JavaScript to the core service, you will not get any compile-time checks, whereas your own service layer would be built in .NET and would therefore have some defences against future (likely) changes in the core service client.
• Consistency with the rest of the CME - In the CME, views are typically considered fully client-side. Where the CME does use Aspx, this is only to generate some HTML on the server, and typically not to for implementation logic.
• Known issues -  ASP.NET postbacks in Anguilla views have been known to cause problems for some people, since e.g. popups won't keep their state through a postback (or an F5 press for that matter).

According to  these criteria, the actual design I was looking at could use the core directly, as my idea was to generate some HTML. In practice, it turns out that there are other reasons to stick with the extra service layer. Even so, I'm very glad I asked the question, and that the answers I got were so informative. Thanks guys!