Welcome to pyrefcount’s documentation!¶
The goal of refcount is to provide a simple API for reference counting, providing protected access to reference
counted values, and for allowing acquire/release actions to be carried out when a reference count becomes active, or
the last user drops off, respectively. refcount is inspired by the Linux Kernel’s refcount_t API.
Motivation¶
While python implements reference counting at the object level through sys.getrefcount(object), there was no
straightforward mechanism for allowing reference counts to be independently defined and controlled.
An example case that motivated this work was a situation where an instance manager class that instantiates and manages multiple subclasses (extended from the same base class) sets a value that only a subset of managed instances make use of, and which modifies the behaviour of the instance manager. These managed instances may be arbitrarily created or deleted, and once the last dependent subclass instance has terminated, the manager wishes to clear the behaviour-modifying value and return to its original behaviour. This particular case involved the computationally expensive conversion of large pandas DataFrames to cuDF ones, which can be skipped when there are no longer any dependent instances that explicitly require cuDF DataFrames to be passed down.