The @atomic decorator now work for class grouped handler

[thomaszdxsn]

I watch aiohttp's doc, it teach me use the aiojobs.atomic to avoid request cancel.

And I found it not work on class grouped handler, I don't mean class view, just this mean method handler:

class UserRelated(object):
     
    def create_user_handler(self, request):
        ...

I investigate the source code, ensure the @atomic decorator cause the problem.

Traceback on class grouped handler

web_protocol.py            310 ERROR    Error handling request
Traceback (most recent call last):
  File ".../lib/python3.6/site-packages/aiohttp/web_protocol.py", line 381, in start
    resp = await self._request_handler(request)
  File ".../lib/python3.6/site-packages/aiohttp/web_app.py", line 322, in _handle
    resp = await handler(request)
TypeError: wrapper() takes 1 positional argument but 2 were given

If you also think this is a bug, I can fix it

[asvetlov]

Sorry, I don't see how to fix it in a generic way. A custom decorator for instance method can be created easily though. Or we can support @atomic(method=True) construction.

[thomaszdxsn]

It is a good idea, but for support back-compat, @atomic must have both call usage and non-call usage.

or just use inspect.ismethod()?

def atomic(coro):
    @wraps(coro)
    async def wrapper(*args):
        if inspect.ismethod(coro):
            instance = args[0]
            if isinstance(instance, View):
                # Class Based View decorated.
                request = instance.request
            else:
                # Method Based View
                request = args[1]
        else:
            request = args[0]


        job = await spawn(request, coro(request))
        return await job.wait()
    return wrapper

but the code look is ugly, and more expensive.

[asvetlov]

It is a good idea, but for support back-compat, @atomic must have both call usage and non-call usage.

Yes.

but the code look is ugly, and more expensive.

That's why I did not support classes from very beginning (and still not sure what way is better now).

[thomaszdxsn]

Ok.

We should postpone this issue if not have better idea.

But maybe we can add some doc for illustrate?

[asvetlov]

Doc example sounds perfect.

[brawer]

Would the approach of https://github.com/nkoshell/aiojobs/commit/20d7a34a74b2fb50a968f875d704ea0279a5ae7e be reasonable? Just wondering because I’ve run into this same problem, but I don’t have enough Python fu to pick up the (apparently abandoned?) work myself.