Corrected special attributes present in object or its direct inheritors by vlasovskikh · Pull Request #874 · python/typeshed

vlasovskikh · 2017-01-27T22:08:00Z

No description provided.

object itself doesn't have __dict__ and __slots__, but all its inheritors do have either __dict__ or __slots__.

…attributes

gvanrossum

Maybe you can separate the non-controversial things out of this so we can merge those separately?

(Also, I realize that PyCharm's needs for typeshed are different from mypy's. Maybe we need to add some way to differentiate, similar to checks for sys.version_info and sys.platform? If you like that idea we should probably discuss this in the typing repo first.)

gvanrossum · 2017-01-30T04:36:44Z

 class object:
 __doc__ = ... # type: Optional[str]
 __class__ = ... # type: type
+ __dict__ = ... # type: Dict[str, Any]


This isn't really correct -- object itself doesn't have __dict__ and neither do classes using __slots__.

gvanrossum · 2017-01-30T04:37:55Z

 def __new__(cls) -> Any: ...
 def __setattr__(self, name: str, value: Any) -> None: ...
- def __eq__(self, o: object) -> bool: ...
- def __ne__(self, o: object) -> bool: ...


Strange you're deleting these, because objectdoes have __eq__ and __ne__.

It's true for type objects like object class, but as for object() instances I'm not sure about it:
>>> object().__eq__ Traceback (most recent call last): ... AttributeError: 'object' object has no attribute '__eq__' >>> object().__ne__ Traceback (most recent call last): ... AttributeError: 'object' object has no attribute '__ne__'

Ah, good catch. Looks like a Python 2/3 compatibility issues -- in Python 3 they exist:
>>> object().__eq__ <method-wrapper '__eq__' of object object at 0x10062e110>

Also, while the methods may not exist, the operations from which they are typically used (== and !=) do work on objects.

(And I just confirmed that mypy fails simple correct code like this:
classA: passA() ==A()
with the error
__tmp__.py:2: error: Unsupported left operand type for == ("A")

gvanrossum · 2017-01-30T04:40:36Z

 def __gt__(self, x: Tuple[_T_co, ...]) -> bool: ...
 def __ge__(self, x: Tuple[_T_co, ...]) -> bool: ...
+ def __eq__(self, x: Tuple[_T_co, ...]) -> bool: ...
+ def __ne__(self, x: Tuple[_T_co, ...]) -> bool: ...


This is also not strictly true -- you can write (1,) == 0 and it'll just return False.
However, there's discussion in python/mypy#1271, so I'm not sure what to do.

Since this issue affects other type checkers as well, not just mypy, I created a typing issue where we can discuss this: python/typing#378

vlasovskikh · 2017-01-30T15:04:38Z

The attributes of object is a special case for a static analysis tool. Both __slots__ and __dict__ are missing in object() instances, but either one or the other is present in the inheritors of object.

I've added __dict__ by analogy to __slots__ that already was here in the stub for object.

Since object is really a special case, we in PyCharm can live without __slots__ and __dict__ in the stub. In that case we will put the knowledge about these attributes into our analysis engine.

Lets either keep or remove both __slots__ and __dict__ (and add special handling for them in our type checkers in the case of removal if needed). @gvanrossum What do you think?

vlasovskikh · 2017-01-30T15:09:05Z

I would try to keep typeshed universally applicable to all the type checkers. At least for now known incompatibility issues seem not that severe to me and I'm looking forward to make our code analysis in respect to understanding typeshed more compatible with PEP 484 and other type checkers. I hope the same applies to other contributors.

JukkaL · 2017-01-30T17:59:09Z

The intention of typeshed is to be useful for all Python type checkers. It's currently somewhat biased towards mypy and pytype, since they have used typeshed for the longest time, but we are happy to discuss the removal of mypyisms etc. These may take a while to be implemented, since they may require changes to mypy first. If we ask you to create a separate issue for something, it doesn't mean that we don't want that change. It may mean that we have to modify mypy before we can merge a part of a PR, since we also want to keep typeshed compatible with tools that already use it.

vlasovskikh · 2017-01-30T18:41:58Z

@JukkaL Yes, sure. I was answering to Guido's comment above about the possibility of adding analyzer-specific checks similar to sys.version_info and sys.platform.

gvanrossum · 2017-01-30T20:09:31Z

For cases where the fix for mypy is difficult or debatable, we could implement a workaround in typeshed using if MYPY -- this works today, as follows:

MYPY=FalseifMYPY:
<versionformypy>else:
<versionforothertype-checkers>

Assuming that other type-checkers treat if MYPY as if False. We should add a comment pointing to the mypy issue that needs to be fixed before the mypy-specific code can be removed.

ambv · 2017-01-30T21:35:39Z

@vlasovskikh If we start sprinkling typeshed with if MYPY:, please contribute some tests we can run against your code. I'm afraid to leave the "non-MYPY"-specific branches in Python 3 type stubs untested. I'd like PyCharm to be more of a first-class citizen in terms of typeshed usage.

vlasovskikh · 2017-01-30T22:22:46Z

Lets at least try to continue without specific checks for type checkers. As we at PyCharm are gradually switching to typeshed we keep finding new PEP 484 incompatibilities in our checker. And we would prefer to fix them rather then add PyCharm-specific checks in stubs.

gvanrossum · 2017-01-30T22:28:14Z

Of course, it's better if we don't have to use checker-specific code in the stubs! Are you okay with the remaining open PRs that are either still under discussion or failing?

ambv · 2017-01-30T22:31:58Z

Agreed, the less checker-specific code, the better. Andrey, maybe there's an easy way for you to provide us with some small checker that we could integrate as part of runtests.sh and Travis CI? That would help ensuring we don't introduce any problems in stubs for PyCharm in the future.

vlasovskikh · 2017-01-31T14:58:53Z

@gvanrossum Yes, I'm OK with that. We'll keep our own set of patches for the open PRs for a while.

vlasovskikh · 2017-01-31T15:01:10Z

@ambv Good idea! We're setting up internal tests for typeshed stubs for PyCharm. I hope we will make them runnable as a part of runtests.sh eventually.

gvanrossum

Again, a suggestion to move forward with the non-controversial changes separately.

gvanrossum · 2017-02-06T18:22:28Z

 def __new__(cls) -> Any: ...
 def __setattr__(self, name: str, value: Any) -> None: ...
- def __eq__(self, o: object) -> bool: ...
- def __ne__(self, o: object) -> bool: ...


Also, while the methods may not exist, the operations from which they are typically used (== and !=) do work on objects.

vlasovskikh · 2017-02-07T01:38:00Z

@gvanrossum Agree, I'll re-send non-controversial bits of this request as another PR.

vlasovskikh added 4 commits January 28, 2017 01:05

Corrected special attributes present in object or its direct inheritors
2beebff
object itself doesn't have __dict__ and __slots__, but all its inheritors do have either __dict__ or __slots__.

Use plain 'str' as type of keys in __dict__ similar to other special …
7514947
…attributes

Removed overridden __dict__ and added object as base class
3c3b402

Added __ne__ and __eq__ to tuple instead of inherited ones from object
706c32a

gvanrossum reviewed Jan 30, 2017
View reviewed changes

JukkaL mentioned this pull request Jan 30, 2017
How should type checkers type check equality? python/typing#378
Closed

gvanrossum reviewed Feb 6, 2017
View reviewed changes

gvanrossum added the blocked label Feb 6, 2017

vlasovskikh closed this Feb 7, 2017

gvanrossum removed the blocked label Feb 13, 2017

vlasovskikh deleted the object-special-attributes branch February 19, 2017 15:52

vlasovskikh mentioned this pull request Feb 19, 2017
Missing special attributes of class instances inherited from object #943
Merged

Uh oh!

Uh oh!

Conversation

vlasovskikh commented Jan 27, 2017

Uh oh!

gvanrossum left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

vlasovskikh commented Jan 30, 2017

Uh oh!

vlasovskikh commented Jan 30, 2017

Uh oh!

JukkaL commented Jan 30, 2017

Uh oh!

vlasovskikh commented Jan 30, 2017

Uh oh!

gvanrossum commented Jan 30, 2017

Uh oh!

ambv commented Jan 30, 2017

Uh oh!

vlasovskikh commented Jan 30, 2017

Uh oh!

gvanrossum commented Jan 30, 2017 via email

Uh oh!

ambv commented Jan 30, 2017

Uh oh!

vlasovskikh commented Jan 31, 2017

Uh oh!

vlasovskikh commented Jan 31, 2017• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

gvanrossum left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

vlasovskikh commented Feb 7, 2017

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

vlasovskikh commented Jan 31, 2017•
edited
Loading