[Opensim-dev] User services refactoring status

[Justin Clark-Casey]

diva at metaverseink.com wrote:
> Tom Willans wrote:
>> Are some aspects of the interface more settled than others? I appreciate 
>> that this is a judgement call.
> 
> Yes, some aspects are more settled than others.
> First, let me point out where this comes from, so that you and others 
> can understand better.
> 
> Up to recently, there was no such thing as OpenSim.Services.Interfaces, 
> nor the companions OpenSim.Services.Connectors and 
> OpenSim.Server.Handlers. The interaction between the simulator and the 
> resource services, and among the resource services themselves, was a 
> mess. I don't mean it in a criticizing attitude, early-stage code is 
> usually a mess. The broad stokes were there, and were really good, but 
> the details were a complete mess. Simple questions like "what handlers 
> are there?" or "what are the resource service interfaces?" were a 
> challenge to answer, interfaces and handlers were spread all over the 
> code, special cases were pervasive, classes were being used directly 
> instead of interfaces, etc. The user services, in particular, were a 
> nightmare, with duplication running rampant because of "standalone vs 
> grid mode", and the code being on the verge of bipolar disorder. Users, 
> agents and avatars were being used interchangeably, with no conceptual 
> distinction, misnomers everywhere, etc.
> 
> This refactoring's main goal is to depart from that chaotic mess and to 
> instill order in the code base. It is primarily a *refactoring*, meaning 
> that code has moved from one namespace to another, but it is also a 
> rewrite -- because in the process of revisiting the code it was clear 
> that things should be done better than they were being done. In 
> particular, harmful hard-coded instantiations of classes were replaced 
> by dynamic loading and instantiation. The core of this is 
> OpenSim.Server.Base.Utils.LoadPlugin -- this is the core of the ROBUST 
> model.
> 
> Having said that, let me now explain how people who plan to extend 
> OpenSim should be looking at it.

Is this documented on the wiki, such as at

http://opensimulator.org/wiki/OpenSim_Services_and_Service_Connectors

?  Unless it's somewhere more consumable than just this mailing list the information is going to get 
lost.

> 
> The main "contract" between the core code and developers of resource 
> services is the collection of interfaces in OpenSim.Services.Interfaces. 
> This is what the simulators call, can't get around hard-coded calls to 
> those interfaces. Those interfaces will stabilize very quickly. Most of 
> them are stable already, with the notable exception of IInventoryService 
> which still needs love. But all others will be stable at 0.7. Now is the 
> right time for people to look at those interfaces and send feedback 
> about which things you think are missing.
> 
> The next level of reuse is the code in OpenSim.Services.Connectors and 
> OpenSim.Server.Handlers. This code is less important than the 
> interfaces, because it's invisible to the simulators, therefore ppl 
> should be more careful about relying on it. This is quite an important 
> point that I don't think many people grasp yet, so let me explain.
> 
> In most systems, the network protocols are fixed. People go to great 
> lengths discussing them ad-nauseum in meetings and committees, to settle 
> on the right data formats, the right data, etc. NOT HERE. The network 
> protocols in OpenSim are now dynamically loaded, and therefore 
> replaceable. There is no such thing as *the* network protocol between 
> the simulator and, say, the asset server. There can be many -- you can 
> roll your own. There are reference implementations in the core 
> distribution, but that's exactly what they are: reference 
> implementations, they are not fixed protocols. As such, people should be 
> careful about relying on those reference implementations.

Reference implementations seems a misnomer here - such code would constitute example implementations 
since they don't act as a reference for a spec.

Anyway, I think that we should be more ambitious than just have example implementations.  In my 
opinion, we should have wire protocol implementations in core that aim for a high level of stability 
but with room for evolution, just as Apache embodied the HTTP procotol which was explicitly designed 
with these properties in mind (as outlined by Fielding in his dissertation about the architecture of 
the web - http://www.ics.uci.edu/~fielding/pubs/dissertation/top.htm).  People who want to 
experiment with other ways of doing things will still be very free to do so.

If we don't have this kind of thing in OpenSim then the interoperability focus will move to 
downstream projects that do attempt this, with all the pros (freedom to narrowly target diverse use 
cases) and cons (fragmentation) that this involves.

> 
> So, to sum it up: the main contract is OpenSim.Services.Interfaces and 
> those are going to be stable very soon. The level right under it, 
> OpenSim.Server.Handlers and OpenSim.Services.Connectors can also be 
> relied upon but it's less stable. I'm still not happy with many of 
> those, so they will change after 0.7. People who don't want to put up 
> with these changes may want to roll their own network protocols.

This sounds like an invitation to fragmentation, which arguably advances some parts of the research 
agenda but penalizes interoperability and network effect gains.

> _______________________________________________
> Opensim-dev mailing list
> Opensim-dev at lists.berlios.de
> https://lists.berlios.de/mailman/listinfo/opensim-dev
> 


-- 
Justin Clark-Casey (justincc)
http://justincc.org
http://twitter.com/justincc