Cannot pickle functions decorated through contextlib

[crusaderky]

It's a somewhat little known feature of contextlib.contextmanager and contextlib.asynccontextmanager that you can use the decorated functions not only as context managers, but also as function decorators:

import asyncio
from contextlib import contextmanager, asynccontextmanager

@contextmanager
def c():
    print("before")
    yield
    print("after")

@c()
def f():
    print("hello world")

f()

@asynccontextmanager
async def ac():
    print("async before")
    yield
    print("async after")

@ac()
async def af():
    print("async hello world")

asyncio.run(af())

before
hello world
after
async before
async hello world
async after

cloudpickle 2.2.1 fails to serialize functions decorated this way:

>>> pickle.loads(pickle.dumps(f))
<function f at 0x7f949972d580>

>>> pickle.loads(pickle.dumps(af))
<function af at 0x7f9499770180>

>>> cloudpickle.dumps(f)
Traceback (most recent call last):
  File "demo.py", line 34, in <module>
    cloudpickle.dumps(f)
          ^^^^^^^^^^^^^^^^^^^^
  File "lib/python3.11/site-packages/cloudpickle/cloudpickle_fast.py", line 73, in dumps
    cp.dump(obj)
  File "lib/python3.11/site-packages/cloudpickle/cloudpickle_fast.py", line 632, in dump
    return Pickler.dump(self, obj)
           ^^^^^^^^^^^^^^^^^^^^^^^
TypeError: cannot pickle 'generator' object

>>> cloudpickle.dumps(af)
Traceback (most recent call last):
  File "demo.py", line 35, in <module>
    cloudpickle.dumps(af)
          ^^^^^^^^^^^^^^^^^^^^^
  File "lib/python3.11/site-packages/cloudpickle/cloudpickle_fast.py", line 73, in dumps
    cp.dump(obj)
  File "lib/python3.11/site-packages/cloudpickle/cloudpickle_fast.py", line 632, in dump
    return Pickler.dump(self, obj)
           ^^^^^^^^^^^^^^^^^^^^^^^
TypeError: cannot pickle 'async_generator' object

[ogrisel]

Thanks for the report, I confirm I can reproduce with master.

The problem is weird because it is raised by the parent class of our pickler which should precisely be the pickle.Pickler class used internally by pickle.dumps...

>>> import io
>>> buffer = io.BytesIO()
>>> pickle.Pickler(buffer, protocol=5).dump(f)
# does not raise