The odyssey of loading a class in a Eclipse plugin – or how we integrated docx4j in POI4XPages

While I was visiting the IBMConnect in Orlando, Lena investigated about how to convert Word files into PDFs. Because of licence reasons, we decide to use docx4j in conjunction with Apache FOP. Both projects are licensed using the Apache V2 license, in opposition to iText.

Lena built a prototype and tested it with Eclipse against the current JVM of the domino server. All worked fine and we started to integrate docx4j as new plugin project in POI4XPages. We also tested the plugin against a regular java programm. Everything worked fine.

The next step was to integrate the creation of the PDF into the UIDocument and the DocumentGeneration classes. This worked fine as well but when we started the conversion it seemed that docx4j has lost its ‘Focus’. The conversion failed because it could not find ‘docx4j.properties’ and ‘log4j.properties’.
After several attempts of building source code based of docx4j and debugging and patching it, I figured out the problem during the night. To make it short, here is the analysis of what had happened:

The problem:

The conversion function in the eclipse plugin was executed by a ClassLoader from Domino. This is not bad at all. But in this case, the context was absolutly relevant to find all the resource files in the docx4j.jar and all the classes. JAXB seems also to be classloading context sensitiv.

The solution:

We changed the ClassLoader context during the PDF conversion as you can see in the code below.
The red marked code does the trick. currentThread.setContextClassLoader( Activator.class.getClassLoader() ) changes the ClassLoader context to the context of the plugin.
When the conversion is done we change the ClassLoader context back to the old context.

public void buildPDF(InputStream isDocument, OutputStream osTarget)
throws PDFException {
Logger logCurrent = LoggerFactory.getLogger(this.getClass().getCanonicalName());
boolean blRC = true;
Exception eRESP = null;
Thread currentThread = Thread.currentThread();
ClassLoader clCurrent = currentThread.getContextClassLoader();
logCurrent.info(“Current thread class loader is: ” +clCurrent);
try {
 currentThread.setContextClassLoader(Activator.class
                    .getClassLoader());
logCurrent.info(“Getting WordprozessingPackage”);
WordprocessingMLPackage wordMLPackage = WordprocessingMLPackage
.load(isDocument);

logCurrent.info(“Getting PdfSettings”);
// 2) Prepare Pdf settings
PdfSettings pdfSettings = new PdfSettings();

// 3) Convert WordprocessingMLPackage to Pdf
logCurrent.info(“Getting PdfConversion”);
PdfConversion converter = new Conversion(wordMLPackage);

logCurrent.info(“do Conversion”);
converter.output(osTarget, pdfSettings);
} catch (Exception e) {
eRESP = e;
logCurrent.log(Level.SEVERE, “Error during PDF Conversion: “+e.getMessage(),e);
blRC = false;
} finally {
            currentThread.setContextClassLoader(clCurrent);
}
if (!blRC) {
throw new PDFException(“Error during FOP PDF Generation”, eRESP);
}
}

A happy ending to a long odyssey.
The new release of POI4XPages will be released later this week, together with some bug fixes and new features.