Cookies for different domains with same name disappear

[JeckLabs]

It seems that cookies are stored in key / value dictionary that causing the problem in the storing cookies from different domains / subdomains in one session.

with aiohttp.ClientSession() as client:

    async with client.get('https://httpbin.org/cookies/set?test=ok') as resp:
        logging.info(await resp.json())

    async with client.get('https://httpbin.org/cookies') as resp:
        logging.info(await resp.json())

    async with client.get('https://http2bin.org/cookies/set?test=fail') as resp:
        logging.info(await resp.json())

    async with client.get('https://httpbin.org/cookies') as resp:
        logging.info(await resp.json())

    logger.info(client.cookies)

2016-08-25 21:57:38 INFO {'cookies': {'test_cookie': 'ok'}}
2016-08-25 21:57:39 INFO {'cookies': {'test_cookie': 'ok'}}
2016-08-25 21:57:39 INFO {'cookies': {'test_cookie': 'fail'}}
2016-08-25 21:57:39 INFO {'cookies': {}}

[iglv]

We are experiencing the same problem.

[JeckLabs]

Some WIP  JeckLabs@2259d07

[asvetlov]

Your WIP hack http.cookiejar.Cookie in ugly way, using very private API.
It's not acceptable.

Another question scaring me is: what to do with ClientSession.cookies?

[JeckLabs]

Hmm, I think I have not used any private API, only public methods, and mock objects. That is the way requests do (see - http://docs.python-requests.org/en/master/_modules/requests/cookies/ ). I agree that's ugly, but allows use standard library code without copy/paste.

I think ClientSession.cookies return a list of Cookie in master. I fix that in JeckLabs@668cbee#diff-7dd84b5ef8d5eea2de1dfc5329411dfcL497 but maybe I missing something.

And now I have problems with tests:

In https://github.com/KeepSafe/aiohttp/blob/master/tests/test_client_functional.py#L1410 setting 3 cookies but expecting only 2.

In https://github.com/KeepSafe/aiohttp/blob/master/tests/test_client_functional.py#L1395 security flag is set but expecting receive cookie through insecure HTTP protocol.

[asvetlov]

I know how requests is implemented and considering it very dirty hack.
Better to copy-paste than mocking.
If aiohttp.CookieJar has bugs -- we should fix them.
Borrowing http.cookiejar implementation ideas is perfectly fine though.

ClientSession.cookies is SimpleCookie actually with dict-like interface.

[JeckLabs]

So I can copy lib from https://hg.python.org/cpython/file/tip/Lib/test/test_http_cookiejar.py rewrite some method, and then copy tests from https://hg.python.org/cpython/file/tip/Lib/test/test_http_cookiejar.py , it's ok? Looks a bit tricky...

[asvetlov]

No, it's not OK.
I've wrote borrow implementation ideas but not "copy-paste the whole documentation along with obsolete modes and unused parts".
Do you want to use cookie policies? Do you want to support netscape-styled cookies? Really?

[asvetlov]

And, please, before starting to hack please tell me what do you want to do with ClientSession.cookies?

[JeckLabs]

I can implement wrapper with SimpleCookie-like interface, it's relatively easy.
Existing aiohttp.CookieJar can be fixed to support multi-domain cookies, but I am don't understand what existed tests do, there may be some more bugs. It seems a lot of work to implement correct cookies support from scratch.

[asvetlov]

SimpleCookie is dict-like object where keys are cookie names.
It's accessible by cookies[name].
But for supporting non-overlapped domains the access should be cookies[domain][name].

[JeckLabs]

Yes, cookies[doman][path][name] to be precise.
For backwards compatibility, I think your can leave SimpleCookie interface for ClientSession.cookies. Advanced users may access cookies structure via some method or additional property.