On Mar 6, 2007, at 11:25 PM, Ben Hood wrote: >> Making a change would be fairly straightforward. It would probably >> be best to change HessianInput/HessianOutput, not the serializers. >> >> We'd also need to add an update to the spec, because technically the >> receiving end is required to keep references to all the objects just >> in case there's a back reference needed. So we'd need to add a >> "streaming hessian" sub-protocol which removed the back references, >> just like we have the "micro-hessian" which skips doubles. > > Cool. I've been demarcating streaming hessian via the content-type > header in Http, so that the receiving application knows its getting a > method call then n objects which it should deserialize and pass to the > method for invocation. This seems like a bit of a hack, and binds the > interaction to using http. I was wondering if their was a more elegant > solution if you were going to introduce a sub-protocol.
We could introduce the streaming call with 'C' x02 x00 instead of 'c' x02 x00. It would probably look like: C x02 x02 H "header" S ... m "method name" data_1 data_2 ... data_n z > Having said that, I wonder if this is going in the direction of an > abstract interaction protocol such as BEEP and beyond the remit of a > pure object serialization protocol. I'd need to look at BEEP again, it's been years since I've looked at it and I can't remember much about it. > I was also thinking that you may not necessarily want to turn back > references off completely, just between the descrete top level objects > in the stream. This may be useful if each object that you are > streaming has a graph that you want use references for, within the > deserialization scope of that object only. So once the nth top level > object has been completely deserialized, then you throw away any back > references it holds. That's a good point. I think the cleanest solution would be to reset the references after each top-level object, i.e. start counting from zero. Anything else would be tricky to specify. -- Scott > > Ben > > _______________________________________________ hessian-interest mailing list [email protected] http://maillist.caucho.com/mailman/listinfo/hessian-interest
