In this issue we found problems with javassisting ops for the purpose of reduction, in that the new op was in a different package (org.scijava) that then couldn't read the source op to directly call the appropriate methods.

The solution involved using reflection instead which has significant performance implications.

We need another round of exploration into ways to make this work.

Also, in the original issue I noted

Forcing the Javassist Op to be in the same net.imagej... package as the original op (FAILED)

This should be better explained exactly what was tried and how it failed, as ideally this would be a great (and sensible) solution.