Code generation for JBJCA

[jeff.zhang]

Hi, I would like to create code generation for JBoss JCA project.

Goal:

To generate code and project configuration of JCA resource adapter. For the user, they can use this tools to generate JCA resource adapter project skeleton. That will include basic resouce adpater, managed connection factory, etc java files and ra.xml, plus ant build.xml file.

User can use this to input some configuration information and generate codes and use ant to package into rar file could be deployed into JCA container.

Research:

So far, it seems no other project could do this job. But there are some projects could generate codes of ejb.

We have 2 subtasks for this:

1. conclusion all important configuration in the rar package, like ra.xml properties, class filename, etc...

2. use template to write down the java source code and resource files

Feature:

Phase 1:

1. use xml file as input file, user can edit this file to define configuration

2. module read from input file and get the variable tree (basic configuration)

3. apply the template to generate files (at first we can use simple write down output)

4. generate build.xml file, user can use ant the compile and package rar file.

5. standalone command line tool and ant task for this

6. JCA 1.6 annotation based and ra.xml defined

7. outbound/NoTx, Inbound&Inout/LocalTx/XA

Phase 2:

1. maven plugin and pom.xml

2. JCA 1.5 and JCA 1.0 profile

3. add more configurations

4. maybe use template thirdparty jar to improve the feature

5. add testsuite for generated code base on JBJCA test framework.

Later:

Maybe eclispe plugin

[jesper.pedersen]

Sounds good

For the first iteration I would just take the input from the command line - like the seam-gen tool. Once we have all the input we need we can then make an XML model of that. That configuration can then be passed into the tool using f.ex. a -f switch.

Create the code under a

codegenerator

modul in the repository, and follow the layout of our current modules.

I think a good first step would be to generate a ResourceAdapter class into a specified output directory. Then we can add more input and generate additional code based on that.

[jesper.pedersen]

A key feature of the tool would be to generate all the boiler plate code for all the config-property's for the various objects, and map them into ra.xml or annotate them. This is really painful work, so lets make it easy.

Of course the skeleton must pass all validation rules of our validation module. So the config-property's comes into play there too for the equals() / hashCode() implementations.

[jesper.pedersen]

Regarding the XML model - I think that using a StAX based model would be the best solution.

See the

javax.xml.stream

package for details - we currently have some code in the repository using this method.

[jeff.zhang]

I commit the initial codes of code generation. That is first step and simple. Just input resource adapter package name and class name and code generator will create a resource adapter java file.

The code is in the trunk\codegenerator.

use "ant sjc" in the trunk directory. It will create jboss-jca-codegenerator.jar in the sjc/lib. Currently we need copy the codegenerator.sh/bat into lib directory and execute it.

Please input ResourceAdapter package name:
org.jboss.jca
Please input ResourceAdapter class name:
BaseResourceAdpater
Java file wrote

Then BaseResourceAdpater.java file output.

PS: Jesper, you said we output the code into doc/codegenerator directory, this is default directory and we can use -output to configure it, correct?

[jesper.pedersen]

Looks good I have placed the code generator in doc/codegenerator in the distribution as we talked about.

Next would be

• Support for adding <config-property> elements - with matching get/set/is methods

• Move text strings in the tool to a resource bundle, so they can be translated (not the generated code)

• Support for adding a header in the generated files
• Basic code formatting rules (indentation, placement of "{", ...)
• Support for an output directory

Also I think it is best to have the input on the same line as the question:

Main resource adapter package: org.jboss.jca.example.jca16annotation

[jeff.zhang]

So far, my idea is based on template engine + variable value. But it seems not easy to implement this feature:

• Basic code formatting rules (indentation, placement of "{", ...)

Using "out.print" from scratch could control these very easy. But if we want replace it by thirdparty template engine jar later, it should be harder.

[jeff.zhang]

An approach is using code formatter tools after code generation.

I found jalopy and eclipse code formatter can do this job well.

But jalopy is big (>1.2M, include swing gui tool) and eclispe code formattor embbedded any eclipse module code, I don't know can we use it in the code.

[jesper.pedersen]

I think you should continue down the path you started - by having special keyword expansion inside on our template model.

Then just have keywords for "  " (indentation - spaces/tabs), "{" (incl. an optional newline character) and so on.

Yes, the templates will be a bit more unreadable to a human, but when we consider some of the requirements down the road - like embedding specification references inside the method definitions - I think that the tradeoff is ok.