[ https://issues.jboss.org/browse/ISPN-7580?page=com.atlassian.jira.plugin.... ] Ramesh Reddy commented on ISPN-7580: ------------------------------------ [~anistor] I am trying to integrate Teiid at a lower level then exposed JAVA Marshaller API level in a nutshell. Current Teiid implementation of JDG translator (with lesser part of Infinispan, as there is no community version), utilizes the JAVA marshall/entity class files created (sometimes generated) for .proto/and annotations, then uses these java class files for CRUD operations. This uses the Query DSL for querying. In the end uses java reflection utilities to scrap the data off the java objects into tabular tuples. With this we have lot of usability issues, in terms whole JAVA entity layer and classloaders and Teiid use of resource adapters and dynamic nature we need for materialization (caching of the table) feature. So, we decided to rewrite this translator again from ground up. In Teiid, we already have metadata layer and query layer. Now, where given the Protobuf definition and portable nature of Infinispan's hotrod protocol, I have converted the a "message" definition in the .proto file into a relational table definition. So, now I can issue CRUD queries against this table, which are against this message. At the same time, using the rest of the information in .proto file about the tags and data type information, I can generate a dynamic Marshaller that represents table encoding/decoding. Now all I need is way to inject this in/out this marshaller when my usecase in play. Thus the required changes to make it SerializationContext generic and ability to replace stock SerializationContext. I did have to duplicate small portions of ProtobufReader/Writer classes in my code, but I do not think they are significant enough for concern for us to think about maintainability in future. I have already written such layer and testing it currently. So, what this does is, we can entirely throw away Teiid's dependence of JAVA definition of entity/marshaller classes. Then on top it, I use the new {{Ickle}} for query layer. Once completed, I can simply connect to the Infinispan cache and read the registered .proto files, generate compatible schema (DDL) for Teiid's VDB, and let user issue SQL queries against it right away. Given a relational schema (DDL), I can even generate (not done yet) an equivalent .proto file and register with Infinispan. Once the remote cache creation feature is available, then I can effectively replace our internal materialization cache with Infinispan implementation as seamless integration. Hope this gives clear context of what I am trying to do. If you are interested, I can show a demo once at a stable state. -- This message was sent by Atlassian JIRA (v7.2.3#72005)