Message5829

I read this bug with interest because any such memory leak bug is critical, and thus blocks 2.5.2. When I first looked at it, I only noticed the mapping of Java classes to Python types in our expose (a relatively small set of classes); however, on second look, it's also of course used by for our general Java proxy support. Hence the concern.

It would seem to me the easiest thing to do is use a WeakHashMap to track this - we only care to keep this around in PyType#class_to_type when there are still objects out there using these classes (as referenced by PyType#underlying_class, or perhaps other references).

I have attached a very small patch that passes the regrtest. Essentially it just does this change in the initialization of the class_to_type map:

class_to_type = Generic.synchronizedWeakHashMap();

Please try this and see if it addresses these memory leaks.