On Thu, Jun 25, 2015 at 12:49 PM, Eric Edmond <t-er...@microsoft.com> wrote: > Hi, > > I have been writing a C++ extension for Python recently, and am currently > fixing the reference counting throughout the extension. As I am very new to > this topic, my question may have a simple answer, but I was unable to find > any mention of the behavior online. > > When using the PyObject_GetItem(obj, key) function, I noticed inconsistent > behavior with various types of obj (for reference, key is always a str > object). When obj is a standard dictionary, the key’s reference count > remains unchanged, as expected. However, when obj is a pandas DataFrame > object, the key’s reference count increases by 2. This could very well be by > design of the DataFrame object doing some internal caching of the string, > but does not appear in the documentation, so I thought I would bring up the > issue.
What is your question? If you want to know why 2 is added to the reference count for pandas objects, the first place to check would be with the pandas developers, as PyObject_GetItem is just going to call the implementation for that type. You can find the CPython code here: https://hg.python.org/cpython/file/9aad116baee8/Objects/abstract.c#l136 -- https://mail.python.org/mailman/listinfo/python-list
