pytorch
5da932ad - Return None correctly from `Tensor.names` (#28659)

Commit
5 years ago
Return None correctly from `Tensor.names` (#28659) Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/28659 Previously, we would return None from `Tensor.names` without bumping the refcount. This is a bug; the Python API requires the developer to increment the refcount on new references to None. This is because None is a singleton object and does not automatically have its reference count bumped when one uses Py_None (which is a pointer to the actual None singleton object). See the following for Python documentation on this: - https://docs.python.org/3/c-api/none.html#c.Py_RETURN_NONE - https://docs.python.org/3/extending/extending.html#back-to-the-example Fixes https://github.com/pytorch/pytorch/issues/28646 Test Plan: - New test. Differential Revision: D18140593 Pulled By: zou3519 fbshipit-source-id: 302a09021b68229e2e7b1b584b3549b30506bdab
Author
Parents
Loading