Protocols

[ilevkivskyi]

Fixes #11

This is a runtime implementation of protocols as per draft PEP 544. (The mypy counterpart is python/mypy#3132)

The implementation is straightforward (mostly I just made current version to behave more similar to __subclasshook__ in collections.abc and added support for __annotations__ in Python 3.6).
Nevertheless, I add a lot of tests. Some of them may look trivial, but previous experience with generics shows that it is better to have them.

There is probably only one corner case worth noting. Consider this situation with for variable annotations:

@runtime
class Proto(Protocol):
    attr: int
class OtherProto(Protocol):
    attr: int
class BadConcrete:
    attr: int
class GoodConcrete:
    attr = 1

Then I propose the following subclass check results:

issubclass(GoodConcrete, Proto) == True
issubclass(BadConcrete, Proto) == False # This is a bit questionable, maybe return True?
issubclass(OtherProto, Proto) == True

The first one is probably obvious, the second one returns False to make this safe:

if issubclass(BadConcrete, Proto):
    BadConcrete().attr

The last one although safe since one cannot instantiate protocols, and this is probably what one would expect.

Attn: @JukkaL @ambv @gvanrossum

[ilevkivskyi]

Something strange happened here (or maybe the comment was deleted), but here is an interesting note to make: because of how NamedTuple behaves at runtime (just making a collectios.namedtuple) it implements corresponding protocols already at the class level, for example:

@runtime
class P(Protocol):
    x: int
class N(NamedTuple):
    x: int

assert issubclass(N, P)  # OK

[Daenyth]

I left a comment that I realized was completely incorrect so I deleted it

[ilevkivskyi]

I will soon make another PR that will add (simplified) Protocol to typing_extensions. We cannot add
all the functionality while Protocol will be in typing_extensions since several classes (like SupportsInt etc.) subclass _Protocol in typing, ideally they should subclass the new Protocol, this is however quite minor thing.

I will keep this PR open (and update depending on the outcome of experiments with typing_extensions).

[gvanrossum]

So IIUC we should first merge #464 and defer this one until the PEP has been accepted, right?

[ilevkivskyi]

So IIUC we should first merge #464 and defer this one until the PEP has been accepted, right?

Yes, also if there will be something we discover while playing with typing_extensions, I will update this PR accordingly (actually there are already few minor tweaks that I need to make here following the review of #464).

[ilevkivskyi]

This was superseded by #649