I _think_ that this has nothing to do with a previously invoked interpreter, but rather something with the way ImageJ handles updates: a new classloader is instantiated, and we require that to be a FijiClassLoader (because ImageJ's PluginClassLoader has no idea about jars/).

If I am correct, then IJ.getClassLoader() will return an instance of PluginClassLoader, not of FijiClassLoader.

You can test this by running this little macro:

     eval("script", "IJ.log('' + IJ.getClassLoader());");

(This requires Java6, because earlier Java requires jars/js.jar, which is not found if the classloader is not a FijiClassLoader).

Clarification: by "updates" I meant "menu updates", AKA Help>Refresh Menus. And such a menu update might be triggered by the Clojure Interpreter.

There is a tiny chance that me recent work on the refresh code helps, if that is not the case, we will have to patch ImageJA a bit more to try to instantiate a FijiClassLoader and fall back to the PluginClassLoader.

I just fixed this bug.

But now I get this instead:

<script>:1: undefined method `include_class' for main:Object (NoMethodError)
...internal jruby stack elided...

This is probably related to the "import all IJ and TrakEM2 and java.lang classes" feature I added recently.

Will keep you posted.

Fixed! It was a matter of importAll() being called too early.