Adrian Seyboldt

Ok, I guess I'm now starting to understand why you prefer to just have tuples, and I can see that in the context of mcbackend that might make a few...

Hm, maybe that was a bit harsh... I actually think mcbackend looks pretty nice, it's just that I'm not so sure it should be the default, and I don't like...

I can do that. It would be nice if we could come up with a better way of passing the function pointer to the julia code, right now I'm using...

This should do the trick: ``` wrap_python_callback = Main.eval( """ using PyCall function wrap_python_callback(callback) py_func_ptr = callback.address func_ptr = convert(Csize_t, py_func_ptr) func_ptr = convert(Ptr{Nothing}, func_ptr) function rhs_wrapper(du, u, p, t)...

Good point about keeping `callback` alive. Unfortunately your simple solution is much slower, in that example it takes 20ms instead of 3.5ms. I really don't know julia well, so maybe...

Is it possible that pyjulia leaks objects that are passed into function in general? I can't get it to finalize an object after passing it to a julia function: ```python...

I opened a separate issue about this leak for pyjulia: https://github.com/JuliaPy/pyjulia/issues/365

Sorry for all the noise, but that leak turned out not to exist :-) I forgot to call the julia gc... But from what I can tell with a little...

This might be relevant: https://optimized-einsum.readthedocs.io/en/stable/backends.html

@brandonwillard I had the same thought at first, but I'm not really sure how this would fit anywhere into either numba_scipy or numba itself. The main complication we have, is...