Re: MCH74A5 - The Java Virtual Machine has ended

Hi Peter,

Typically, when I receive MCH74A5, it's because the RPG code failed to properly handle a prior error.

For example, I might call a Java routine that opens a file and uses the file to populate an object, then return that object to my RPG code. The object might then be passed to other methods to do some processing.

What happens if the open of the file failed? The Java routine may have returned a null Java object. Or it may have sent an exception back to my program, and I may have chosed to MONITOR for the exception and keep my program running. In either case, I now have a null Java object.

If I decide to continue and try to pass my null object to the other methods that are supposed to process it, I may very well get an MCH74A5 because my code didn't stop the program and/or handle the error appropriately. I tried to proceed, even though I didn't have a valid object, and that caused Java to end in confusion.

I haven't worked with Xalan, so I might be off-base. But I would suggest watching your program very carefully and looking for situations where an error might occur that you might be ignoring. If you add code to handle it properly, you may be able to prevent the JVM from ending.



Peter Connell wrote:

What seemed a simple java from RPG solution prove to be my undoing.

The XALAN class is called from RPG to transform XML into HTML using a
custom XSL style-sheet.

This works fine until the style-sheet is found to contain some syntax
that causes XALAN to cough up an error which kills the JVM resulting in
MCH74A5.

The problem being that once MCH74A5 has occurred then any subsequent
attempts to use a JVM fail irrespective of which RPG program from which
they originate. So the job must be ended, in this case an HTTP server
thread.

I'm hoping that this situation does not occur in production since the
XSL style-sheet transformation will have been tested before promotion.

But it is a bit of a worry that java can potentially cause a whole job
be terminated instead of just the last execution stack.

Even so, it's still a bummer to have to end the HTTP job even during
development while iterating through style-sheet enhancements.

I don't seem to see any way around this.

Peter