Looking through the repo, one change of note that doesn't seem to be documented in the README [0] is that this has an optimization to reduce Copy on Writes in forked workers (under the Py_IMMORTAL_INSTANCES preprocessor guard) by not dirtying pages via refcount changes.
Here's the BPO issue [1] that tracks the upstreaming attempt.
It’s also likely that this change, which is a big win for our prefork production environment, drags down our benchmark results. It makes every incref and decref more expensive.
Here's the BPO issue [1] that tracks the upstreaming attempt.
[0] https://github.com/facebookincubator/cinder/blob/9de726349eb...
[1] https://bugs.python.org/issue40255