Message5831

I tried this out and it didn't work. After applying the patch I created a test app that recreates the problem: it is a java program that creates a PythonInterpreter that runs a python script which contains a python class that implements a java interface. After running the the script I print out the size of class_to_type, do an explicit System.gc(), and print out the size again. This whole process it done in a loop, and the output indicates that the size of class_to_type increases by one on each iteration.

I looked at the source again and I think I know why this didn't work: The weak hash map only contains weak references to the *key*, not the value (it contains a normal strong reference to the value.) The javadocs for WeakHashMap indicate that if the value has a strong reference to the key then the weak hash map will never release that entry, because the (key has a strong reference through the value back to the gc root.) This is exactly the case here: the PyJavaType value contains a reference to the key in its member variable "javaProxy" (you can see this assignment in PyJavaType.java:203)