Issue with aiohttp + vcr: multidict 4.7.0 might be an issue

[sudodoki]

Hello. Thanks for this awesome tool.
On dec 11, 2019, multidict released a new version and it seemed to break our CI setup. We are using aiohttp (3.6.2) + vcrpy (2.0.1) on Python 3
Freezing multidict==4.6.1 seemed to fix the issue. And even though it might be something related to changes in aiohttp, just leaving it here in case anyone else runs into this.

[sudodoki]

If I get time (which I might not) gonna open a corresponding issue in aiohttp. Closing this, as it's probably not directly a vcr issue

[sudodoki]

For SEO related purposes, I was getting
aiohttp.client_exceptions.ClientConnectorError: Cannot connect to host ... which was expected as host was a nonexistent one, but meant that requests were bypassing vcr and trying to go straight for upstream

[neozenith]

Thanks for reporting and documenting this issue.

[vEpiphyte]

@sudodoki @neozenith This was a regression in multidict which should be fixed soon.

aio-libs/multidict#418

[sudodoki]

Seems that 4.7.1 was released and issue might be fixed, but not sure, how to handle that in library's dependencies to avoid accidentally getting the one that doesn't work

[neozenith]

We can put in our setup.py some multidict!=x.y.z specifiers.

https://setuptools.readthedocs.io/en/latest/setuptools.html#declaring-dependencies

Just need to know which versions or a range we should avoid.

[sudodoki]

4.6.1 worked ok, 4.7.0 (following version) - didn't work, 4.7.1 (newer one, supposedly, the one fixing the regression) didn't work for me as well (I didn't rewrite the cassettes though).
I use match_on=["method", "path", "query"] in my VCR constructor, not sure, whether there is any logiс in regards to comparing multidicts or something like this.

[vEpiphyte]

@sudodoki Can you document how your service is breaking with the updated multidict? I'm seeing successful tests using aiothttp + 4.7.1.

[sudodoki]

@vEpiphyte Yeah, I can try, but it might be the issue with me updating multidict only and using it with older aiohttp

[vEpiphyte]

@sudodoki I reproduced an issue issue when working with a project that had some older cassettes which had yaml pyobject hinting in them (vcrpy 1.13 and multidict 4.7.1). Ultimately the deserializer code was hiding the following exception:

code: https://github.com/kevin1024/vcrpy/blob/master/vcr/serialize.py#L40

    try:
        data = serializer.deserialize(cassette_string)
    # Old cassettes used to use yaml object thingy so I have to
    # check for some fairly stupid exceptions here
    except (ImportError, yaml.constructor.ConstructorError):
        _warn_about_old_cassette_format()   <---------------------- Hides the exception
    if _looks_like_an_old_cassette(data):
        _warn_about_old_cassette_format()

I encountered the following exception:

Traceback (most recent call last):
  File "/home/user/.pyenv/versions/3.7.3/envs/someproject373/lib/python3.7/site-packages/yaml/constructor.py", line 532, in find_python_name
    __import__(module_name)
ModuleNotFoundError: No module named 'multidict._istr'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/home/user/.pyenv/versions/3.7.3/envs/someproject373/lib/python3.7/site-packages/vcr/serialize.py", line 36, in deserialize
    data = serializer.deserialize(cassette_string)
  File "/home/user/.pyenv/versions/3.7.3/envs/someproject373/lib/python3.7/site-packages/vcr/serializers/yamlserializer.py", line 11, in deserialize
    return yaml.load(cassette_string, Loader=Loader)
  File "/home/user/.pyenv/versions/3.7.3/envs/someproject373/lib/python3.7/site-packages/yaml/__init__.py", line 114, in load
    return loader.get_single_	data()
  File "/home/user/.pyenv/versions/3.7.3/envs/someproject373/lib/python3.7/site-packages/yaml/constructor.py", line 43, in get_single_data
    return self.construct_document(node)
  File "/home/user/.pyenv/versions/3.7.3/envs/someproject373/lib/python3.7/site-packages/yaml/constructor.py", line 52, in construct_document
    for dummy in generator:
  File "/home/user/.pyenv/versions/3.7.3/envs/someproject373/lib/python3.7/site-packages/yaml/constructor.py", line 404, in construct_yaml_map
    value = self.construct_mapping(node)
  File "/home/user/.pyenv/versions/3.7.3/envs/someproject373/lib/python3.7/site-packages/yaml/constructor.py", line 210, in construct_mapping
    return super().construct_mapping(node, deep=deep)
  File "/home/user/.pyenv/versions/3.7.3/envs/someproject373/lib/python3.7/site-packages/yaml/constructor.py", line 131, in construct_mapping
    key = self.construct_object(key_node, deep=deep)
  File "/home/user/.pyenv/versions/3.7.3/envs/someproject373/lib/python3.7/site-packages/yaml/constructor.py", line 94, in construct_object
    data = constructor(self, tag_suffix, node)
  File "/home/user/.pyenv/versions/3.7.3/envs/someproject373/lib/python3.7/site-packages/yaml/constructor.py", line 635, in construct_python_object_new
    return self.construct_python_object_apply(suffix, node, newobj=True)
  File "/home/user/.pyenv/versions/3.7.3/envs/someproject373/lib/python3.7/site-packages/yaml/constructor.py", line 624, in construct_python_object_apply
    instance = self.make_python_instance(suffix, node, args, kwds, newobj)
  File "/home/user/.pyenv/versions/3.7.3/envs/someproject373/lib/python3.7/site-packages/yaml/constructor.py", line 715, in make_python_instance
    suffix, node, args, kwds, newobj, unsafe=True)
  File "/home/user/.pyenv/versions/3.7.3/envs/someproject373/lib/python3.7/site-packages/yaml/constructor.py", line 566, in make_python_instance
    cls = self.find_python_name(suffix, node.start_mark)
  File "/home/user/.pyenv/versions/3.7.3/envs/someproject373/lib/python3.7/site-packages/yaml/constructor.py", line 711, in find_python_name
    return super(UnsafeConstructor, self).find_python_name(name, mark, unsafe=True)
  File "/home/user/.pyenv/versions/3.7.3/envs/someproject373/lib/python3.7/site-packages/yaml/constructor.py", line 535, in find_python_name
    "cannot find module %r (%s)" % (module_name, exc), mark)
yaml.constructor.ConstructorError: while constructing a Python object
cannot find module 'multidict._istr' (No module named 'multidict._istr')
  in "<unicode string>", line 48, column 9

This causes deserialization failures and subsequently the cassette appears empty and vcrpy has no content to replay. This can result in a existing cassette being overwritten. Reverting to multidict 4.6.1 works.

Going to try re-recording the cassettes with a newer version of VCRPy to see if this issues persists or is just an artifact of the yaml serialization including the pyobject reference.

[vEpiphyte]

Using VCRPY 3.0.0 and multidict 4.7.1 works fine. The object serialization was fixed in the 2.1.x branch due to #454 .