difference in calling conventions leads to strange behavior in pandas about libpython-clj HOT 4 CLOSED

@cnuernber one option -- not a holistic fix, but a fix that would handle most use cases:

I can retool the functions/methods populated by require-python to use (py/call-kw ...). This would still lead to some inconsistencies if users access the lower level raw Python function, but I think it would handle the majority of use cases that users see. The drawback would be a slight hit to performance (an additional nested function call) and the function employed would no longer be a true Python function but a Python function wrapped with a clojure function that calls py/call-kw.

from libpython-clj.

I would like to take a careful look at this. Everything filters through py/call-kw eventually. The questions I have right now are twofold:

1. Is this because of the marshalling layer on top of functions that attempts to ensure we don't deep-copy objects but we do shallow copy simple objects?
2. Is this because the system to detect kwargs via detecting keywords in the arguments?

Basically, I want to know the exact mechanism of failure which could probably be found by printing out the arguments to PyObject_Call xxx.

from libpython-clj.

Closing this, it works with the referenced changelist.

from libpython-clj.

Screencast: https://drive.google.com/file/d/1PTXzWqNaRAiIDDZWqkeffIK2KESRWSRh/view?usp=sharing

from libpython-clj.