add support for multi-phase processing and processor orchestration

[slipperyseal]

Hi everyone,

i'v started writing a transcoder using spoon and I'v based my code on the Launcher example which uses a Factory/ProcessingManager (see below) which then sends CtClass and CtInterface instances to two AbstractProcessors. My question is around the best way to approach multi-phase processing. I can happily extract all the class information from CtClass/CtInterface but I find I'm going to need to scan source tree twice. Scan one to do analysis and the second scan when I generate the output.

I guess my question is what the recommended approach for this is. If I simply keep my own references to all the CtClass/CtInterface instances could this cause memory issues, having all nodes is loaded in memory. Does the processor throw away each class as it moves to the next?

Or, do I just run two of these?

        final String[] args = { "--input", path, "--output-type", "nooutput" };
        launcher.setArgs(args);
        launcher.run();

        runProcessingManager();
        runProcessingManager();
    }

    private void runProcessingManager() {
        Factory factory = launcher.getFactory();
        ProcessingManager processingManager = new QueueProcessingManager(factory);
        processingManager.addProcessor(new ClassProcessor());
        processingManager.addProcessor(new InterfaceProcessor());
        processingManager.process(factory.Class().getAll());
    }

[surli]

Hi @slipperyseal

maybe I'm totally wrong - I'm still a newbie with using Spoon - but why don't you override the method isToBeProcessed in your processors (see AbstractProcessor) to match the element you want to process and then add directly your processor to the launcher using launcher.addProcessor(...)? In the launcher, the model will be built and then each processor will be launched on it, avoiding you to call the process method...

[pvojtechovsky]

If I simply keep my own references to all the CtClass/CtInterface instances could this cause memory issues, having all nodes is loaded in memory. Does the processor throw away each class as it moves to the next?

Whole spoon model is already loaded in memory and stays there. So the collecting of the list of references to the existing spoon model elements should not be a problem.

[slipperyseal]

hi @surli - thanks, yeah i was already using isToBeProcessed to have it call process. was just trying to determine the best method for a multi-pass approach.

thanks @pvojtechovsky - i guess that makes sense that all the inter-class references are loaded up into one big model. I'll run the processor once, keep the references and run my own phases across those.

[monperrus]

specifying a list of processors is the only way to do it declaratively. otherwise, you have to write the multiple phases imperatively.

this discussion seems related to point "Processor orchestration" of the roadmap.

[slipperyseal]

thanks everyone. I noticed that the Processor is simply fed the type list from factory.Class().getAll() which is all I needed anyway.

I've started writing a Java to C++ transcompiler. I'm probably not writing it in the best way (yet), but I've already got basic class structure, abstract methods and interfaces and simple assignment and return values for limited test cases.

https://github.com/slipperyseal/trebuchet

Happy new year!

[monperrus]

https://github.com/slipperyseal/trebuchet

really cool!

[monperrus]

see 'Source model analysis using the JJTraveler visitor combinator framework'