Saxon

Dear all,

I evaluate the EE edition currently, and have some questions. I hope you can verify/correct my assumptions.

Scenario: Process large XML Documents (up to 5 GB) with XSLT. Idea is to take advantage of the streaming features.

First, this is the JAVA code I currently use for testing:

		EnterpriseConfiguration conf = new EnterpriseConfiguration();
        conf.setConfigurationProperty(FeatureKeys.LICENSE_FILE_LOCATION, "M:\\workspace\\cau\\PoC_MM_2\\libs\\saxon-license.lic");
        conf.setConfigurationProperty(FeatureKeys.STREAMABILITY, "standard");
        //conf.setConfigurationProperty(FeatureKeys.TREE_MODEL_NAME, "tinyTreeCondensed");
        log.info(conf.getEditionCode());


        Processor entProc = new Processor(conf); // Should be created only once
        log.info(entProc.getUnderlyingConfiguration().getEditionCode()); // Shall be EE
        XsltCompiler comp = entProc.newXsltCompiler();
        XsltExecutable exp = comp.compile(new StreamSource(new FileInputStream("C:\\Temp\\dummyCleared.xslt")));
        Xslt30Transformer xslt30Transformer = exp.load30(); // Should be created for each transformation
        FileInputStream fis = new FileInputStream("C:\\Temp\\dummyCleared.xml");
        StreamSource ss = new StreamSource(fis);
        FileOutputStream fos = new FileOutputStream("C:\\Temp\\resultCleared.txt");
        Serializer out = entProc.newSerializer(fos);

        xslt30Transformer.applyTemplates(ss, out);
        xslt30Transformer.getUnderlyingController().clearDocumentPool();
        out.close();
        fos.close();
        fis.close();

The test XML:

			
			
			
				
					
						
							XYZ001375-2030-00-0002-1009600
							XYZ001375-2030-00-0002-1009600
						
					
				
				
					
						
							loc3245
							
						
					
				
			

And this is a basic test XSLT:

Now my questions:

1. Even if it seems to stream the content, a 1,3GB file is still using 3.5 to 4 GB of RAM - is this normal?
2. After usage, the memory is not cleared - the process is not freeing it. Am I missing something in the code?
3. Does the overall approach seem to be feasible? I would have expected less memory consumption - am I wrong there?

Thanks a lot and warmest regards, Manfred