Static Files Error With The Latest Version (3.3.2)

[GitBolt]

There seems to be something wrong with serving static files in the latest version, this is the error:

raise TypeError("async_to_sync can only be applied to async functions.")
TypeError: async_to_sync can only be applied to async functions.
HTTP GET /static/rest_framework/js/default.js 500 [1.01, 127.0.0.1:50017]

The previous version (3.3.1) works perfectly

[carltongibson]

@StingBolt Can you paste the full traceback please? Also can you say if you're using this just with Django or with Channels or with..., and so on? — Context is all here. 🙂

Thanks

Related django/django#14221

[BeryJu]

I also just ran into this issue, same error message:

Traceback (most recent call last):
  File "/Users/langhammerj/.local/share/virtualenvs/authentik-fAhVAmau/lib/python3.9/site-packages/channels/staticfiles.py", line 40, in __call__
    return await self.staticfiles_handler_class()(
  File "/Users/langhammerj/.local/share/virtualenvs/authentik-fAhVAmau/lib/python3.9/site-packages/channels/staticfiles.py", line 56, in __call__
    return await super().__call__(scope, receive, send)
  File "/Users/langhammerj/.local/share/virtualenvs/authentik-fAhVAmau/lib/python3.9/site-packages/channels/http.py", line 198, in __call__
    await self.handle(scope, async_to_sync(send), body_stream)
  File "/Users/langhammerj/.local/share/virtualenvs/authentik-fAhVAmau/lib/python3.9/site-packages/asgiref/sync.py", line 105, in __init__
    raise TypeError("async_to_sync can only be applied to async functions.")
TypeError: async_to_sync can only be applied to async functions.

(In my case, using Django 3.1.7, asgiref 3.3.2, python 3.9.2 and channels 3.0.3, and downgrading to 3.3.1 fixes it here too)

Oh, and the error also happens if I disable channels and only use django's get_asgi_application()

[KristobalJunta]

I'm not familiar with REST framework code there, but the issue can be related to how functools.partial works on Python < 3.8.

See this SO answer for details - https://stackoverflow.com/a/52422903.

Got a related bug in another library today - datadvance/DjangoChannelsGraphqlWs#68

[pgjones]

I've seen this bug in Quart (I think), here is a fix

[yedpodtrzitko]

Quart code:

 if sys.version_info >= (3, 8):
        return asyncio.iscoroutinefunction(func)

...however I'm encountering this issue even with Python 3.9.1

[andrewgodwin]

The issue is fixed on main, I'll get a 3.3.3 release out today as well.

[jaknel]

This bug is not resolved for me despite upgrading to 3.3.3.

Python 3.8.6
Django 3.2
Channels 3.0.3
asgiref 3.3.3
daphne 3.0.1

It looks like a lambda created by daphne at https://github.com/django/daphne/blob/main/daphne/server.py#L206 is what eventually ends up in https://github.com/django/asgiref/blob/main/asgiref/sync.py#L120 as awaitable and fails inspect._has_code_flag(awaitable, inspect.CO_COROUTINE) which is what _iscoroutinefunction_or_partial(awaitable) eventually calls

[andrewgodwin]

Yup, I think you're right. I'm going to go change the error to a warning for now, async functions are too hard to reliably detect in Python.

[andrewgodwin]

3.3.4 is out which replaces this error with a warning, so your applications will still run. We'll have to figure out how to get that Daphne lambda to "look async", or maybe just abandon the whole idea of detecting coroutine-yielding callables before the fact entirely.

[carltongibson]

Hey @andrewgodwin — thanks for your work on this!

This all went very quickly yesterday, and with the Django 3.2 release I didn't have the bandwidth to follow totally. I was looking at a quick release for Daphne today, but with asgiref 3.3.4 out, I will take a few days to fully work through it, and look to release next week.

We'll have to figure out how to get that Daphne lambda to "look async"...

We have django/daphne#364 proposed, to use a partial instead of the lambda.

[andrewgodwin]

Yes, let's take the time and do this properly. Given we know that partial works and is accounted for in both python and asgiref, I think I'd go with that?