Her first instinct was brute force. She tried to run the JNLP with OpenWebStart, an open-source resurrection of Java Web Start. It failed. The old JARs had dependencies on Sun's proprietary imaging libraries that no longer existed. She tried to decompile the actuarial-core.jar using JD-GUI. The code looked like it had been written by a traumatized Perl programmer in 2005: Vector v = new Vector(); Enumeration e = v.elements(); No comments. Method names like doTheThing() .
The next morning, she deployed her solution as a scheduled Lambda function. At 4:00 AM, the old cron job tried to run the JNLP and failed. But at 4:01 AM, her Lambda woke up, called the SOAP service, ran the Java bridge, and deposited a pristine PDF in a compliance bucket. She even added a small script that emailed Gerald: "Loss run report ready. (Legacy conversion successful.)"
<?xml version="1.0" encoding="UTF-8"?> <jnlp spec="1.0+" codebase="http://legacy-box:8080/actuarial/"> <information> <title>Loss Run Generator</title> <vendor>GIC Legacy Systems</vendor> </information> <resources> <j2se version="1.6+" java-vm-args="-Xmx512m -XX:PermSize=128m"/> <jar href="actuarial-core.jar" main="true"/> <jar href="pdf-generator-2009.jar"/> <jar href="apache-xerces-2.9.1.jar"/> <jar href="jai-core-1.1.3.jar"/> </resources> <application-desc main-class="com.gic.legacy.LossRunMain"/> </jnlp> "Convert JNLP to PDF," she muttered, tasting the absurdity. It was like saying "convert a car engine to a croissant." One was a deployment descriptor for old Java applications. The other was a document format. But the business need was real: inside that JNLP was the recipe for a PDF. And she needed to extract it. convert jnlp to pdf
Now the challenge: she needed to "convert JNLP to PDF" in a way that was automated, serverless, and modern. She couldn't run the JNLP anymore. But she could extract its soul.
# convert_jnlp_to_pdf.py - Elena Vasquez, 2024 # Takes a JNLP file path, extracts resources, builds bridge, outputs PDF. # R.I.P. Java Web Start. You were annoying, but you did your job. And somewhere in the cloud, at that exact moment, her Lambda ran again, producing a PDF that would be printed, signed, and filed with a state commission that had no idea their insurance reports owed their existence to a twenty-year-old JNLP file and a woman who refused to say "it's impossible." Her first instinct was brute force
Then, the application would take that XML, run it through a series of XSLT transformations (the apache-xerces JAR), feed the result into the pdf-generator-2009.jar (which was a thin wrapper over iText 2.1.7, a version so old it predated PDF/A standards), and finally spit out a byte array that was written to C:\legacy_reports\output.pdf .
That night, Elena sat in her home office, the glow of three monitors painting her face blue. She opened the JNLP again. Not as code, but as a puzzle. The JNLP wasn't the enemy. It was a map. It pointed to resources, to a main class, to JVM arguments. The PDF was in there somewhere. The old JARs had dependencies on Sun's proprietary
But Elena knew the truth: she hadn't "converted JNLP to PDF." She had reverse-engineered a zombie. The JNLP was still dead. But its brain—the transformation logic, the JARs, the XSLT—was now puppeted by modern code.
She wrote a Python script. Not because Python was the best tool, but because it was the most forgiving. She used subprocess to call a small Java CLI tool she wrote in twenty minutes. That Java tool did only one thing: loaded the old JARs (including iText 2.1.7 and the Xerces XML parser) into a custom classloader, instantiated the PDFBuilder class via reflection, and exposed a simple method: byte[] generatePdf(String xmlData) .
Elena stared at the .jnlp file in Notepad++. It looked like an alien artifact: