email.message_from_bytes etc return legacy Message objects, not EmailMessage

[fsc-eriker]

When you run MyPy on code which requires an EmailMessage, you often get errors.

from email import message_from_binary_file
from email.message import EmailMessage
from email.policy import default

def fn(msg: EmailMessage) -> None:
    print(type(msg))

with open("/tmp/message.eml", "rb") as emlbytes:
    message = message_from_binary_file(emlbytes, policy=default)
fn(message)

Running mypy on this file gets me

bash$ mypy /tmp/mypy.py 
/tmp/mypy.py:10: error: Argument 1 to "fn" has incompatible type "Message"; expected "EmailMessage"  [arg-type]
Found 1 error in 1 file (checked 1 source file)

You can obviously work around this with a cast or etc; but perhaps the default for these methods should be changed to return an EmailMessage, and require a cast if you need the legacy Message type instead?

I get similar behavior with email_from_bytes etc; all of these functions for producing an EmailMessage object from the RFC5322 representation of an actual message have an API which (obscurely!) depends on the presence of the policy= argument for whether to return an EmailMessage or a legacy Message.

[fsc-eriker]

I don't seem to be able to control the labels myself; please add the "stdlib" and "email-topic" labels, as well as perhaps "topic-typing".

[srittau]

This also applies to the return types of the different email.parser.Parser classes.

We should consider making the Parser classes and the functions generic, similar to FeedParser. For this to work, we would also need to make Policy generic over the message class. Any PRs welcome!

[AlexWaygood]

(FYI, @fsc-eriker, I transferred this issue from CPython since CPython itself has no type hints; all type hints for the standard library are found here, in the typeshed repo :)

[srittau]

#10769 is a first step by making policies generic.