Part of the problem is that JavaMetaClass.new assumes that it will always be called to create a proxy, when it will be called for every descendant class. On the same line, handling of the jclass_registry must be restricted to actual proxies. Something else I found is that the treatment of objects returned from Java is lax, which makes it impossible to pass them back to Java methods.
I can work on some of this issues in due time.
I very much like how fast jnius is at loading Java, and being able to use the latest versions of Python.
Want to back this issue? Post a bounty on it! We accept bounties via Bountysource.
Sorry for the delay, can you provide more specific (testcases?) description for each of those issues? I started some work to allow inheritance of java classes from python (using proxies), but it's not complete, however, the issues you point here seems loosely related to that, and should probably be splitted in several different issues.
Sorry to come into the discussion, but inheriting Java a class might not be what you need @Apalala, I think that it's a better idea to wrap the Java class in a Python class like it's done in this class. It might not be obvious in the linked code because all the autoclass are done in java.py but the code is equivalent to
class MyJavaWrapperClass(AnyPythonClassOrMixin):
def __init__(self, *args, **kwargs):
JavaClass = autoclass('org.uber.cool.JavaClass`)
self._my_java_object = JavaClass(*args, **kwargs) # use double underscore to "hide" the Java object
The reasons not to inherit the Java class are:
class Element
, class Node
and class Relationship
in graph almost all of them has boilerplate code even if most of the time it is to make the API more Pythonic ;)From a performance perceptive, I don't think there is any impact but @tshirtman might know better.
Another thing, wrapping Java objects works right now and really well :-)
Is overriding java method using python code possible for now?
What about java class inheritance?
I've spent quite a lot of time looking in this, and i had come to the conclusion that there was no practical way of doing so, short of generating java bytecode at runtime, there are libraries to do that, but that felt like adding a lot of complication to the library, and probably out of scope. Not sure about closing, but i don't think it's going to get worked on, so it wouldn't be unreasonable to mark it as wontfix.
@monami7001 any progress?
that was ~4 years ago, and to my knowledge no, generating java bytecode at runtime still looks like the only option (and maybe it's not as bad as i though and we should look into that).
One workaround is:
to use it with autoclass