How does NumPy handle internal memory ownership tracking in ndarrays?

NumPy tracks memory ownership using base pointers and reference counters to manage shared buffers.

Each ndarray contains a pointer to its data buffer and a base attribute that references the original owning object. If an array is a view, it does not own memory and instead references the base object. Reference counting ensures that memory is freed only when all views and owners are released. This mechanism prevents premature deallocation and enables safe zero-copy operations.