instead, try ice library from: .zeroc.
it is: free parallel and cross platform.

We have a large hybrid desktop app (Java and DotNet) and marshal xmlized business objects through the bridge bidirectionally. Since it’s all handled between processes on the client machine, the overhead of the bytestream version of the object is nil and response time is instantaneous.

There are many ways this bridging can be deployed but we’ve chosen serializarion via xml so that the bridge is only used for native data types like string or long. JA.Net is also capable of serializing and unserializing data structures, but there’s more overhead with complex recursive object graphs if you approach the problem generically as they would have to do.

- Peter

What I could do with is CORBA, which is as easy to use, understand and administer as WebServices with similar quality of tool support.

CORBA did an excellent job of the mapping, all the way from endieneness to logical Object abstractions. However, as anyone who has ever had to use CORBA in anger will tell you the tools were terrible, the vendors runtimes buggy and in an effort to universally solve this mapping problem it was hugely complex.

WebServices have, through good tool support and the simplicity of the underlying transport (I think most people when they got into webservices did it over http, even if these days people are putting more envelopes on message buses), taken off, people use them. But the simplicity and clarity is an overhead, run a webservice endpoint and profile how much time is being spent marshalling and demarshalling the XML. For some applications I’ve worked on it’s crippling, IBM buying DataPower was a shrewd move.

If you want to have a go at building a simple interop library, port Java DataInput/OutputStreams to C# (the source is now GPLd), that takes about 20 minutes. Then decide that you order your attributes alphabetically and write them to the stream, remember you have a way of sending all the primitive structures (and Unicode strings). At this point there is nothing stopping you from reading this data back into a set of C structs, if you understand the order of the stream. In your OO language of choice the next thing is identify the object you rehydrate your serialized representations into, use a class name, a version number, whatever you like.

There are standards for bytes which you can layer ontop of these things on a case by case basis, need IEEE-754 number representations? Use it.

Last but not least, throw in a code generator, (have a look at GRAM) to get from one source representation to another.

Job, as they say, is a good’en.