Support metaclass subclasses

[mbustosorg]

Descendants should check if the class is a Metaclass. Added test also.

[kernc]

Where is d.py (D) tested for?

[kernc]

Where is d.py (D) tested for?

[mbustosorg]

d.py is tested as part of the example_pkg test. Calling with self.obj doesn't work because non metaclass type classes take no arguments. The tests fail that way. Thanks, mauricio

…

On Mon, Apr 1, 2019 at 5:36 AM kernc ***@***.***> wrote: ***@***.**** commented on this pull request. Where is *d.py* (D) tested for? ------------------------------ In pdoc/__init__.py <#48 (comment)>: > @@ -1210,6 +1210,10 @@ def subclasses(self) -> List['Class']: The objects in the list are of type `pdoc.Class` if available, and `pdoc.External` otherwise. """ + if issubclass(self.obj, type): + return [self.module.find_class(c) + for c in self.obj.__subclasses__(self.obj)] + I think the whole method can be replaced with: return [self.module.find_class(c) for c in type.__subclasses__(self.obj)] This makes the branch redundant as it works for both cases. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#48 (review)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAt-cWy7IPNj1NUngbyZGDvkG9zAElU4ks5vcf03gaJpZM4cUuQK> .

[kernc]

I think that's because self.obj.__subclasses__ is a bound method whereas type.__subclasses__ is unbound and can be passed the self.obj (meta)class parameter. Seems to work.

[mbustosorg]

I tried that iteration and it failed when testing d.py. Did you see different results?

[kernc]

Seems to work here on Python 3.5:

>>> class A: pass
>>> class B(type): pass
>>> class C(A): pass
>>> class D(B): pass
>>> type.__subclasses__(A)
[<class '__main__.C'>]
>>> type.__subclasses__(B)
[<class '__main__.D'>]

In addition, would you consider creating a new test ApiTest.test_subclasses() where you create a few local classes and call pdoc.Class(cls.__name__, sys, cls).subclasses() on them?

[mbustosorg]

Yes. I'll have a look at why I'm seeing different results as you and will create the new test case.

…

On Mon, Apr 1, 2019 at 6:30 PM kernc ***@***.***> wrote: Seems to work here: >>> class A: pass>>> class B(type): pass>>> class C(A): pass>>> class D(B): pass>>> type.__subclasses__(A) [<class '__main__.C'>]>>> type.__subclasses__(B) [<class '__main__.D'>] In addition, would you consider creating a new test ApiTest.test_subclasses() where you create a few local classes and call pdoc.Class(cls.__name__, sys, cls).subclasses() on them? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#48 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAt-cU1BRvSFb8Hj2VWliuuXVjXAZFLeks5vcrKzgaJpZM4cUuQK> .

[mbustosorg]

So you would recommend creating a new submodule that includes these classes instead of in the code like this?

…

On Tue, Apr 2, 2019 at 3:18 PM kernc ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In pdoc/test/example_pkg/_test_linking/d.py <#48 (comment)>: > @@ -0,0 +1,10 @@ +class D(type): + + def __init__(cls, name, bases, dct): + super().__init__(cls, name, bases, dct) + cls._instance = None + + def __call__(cls, *args, **kwargs): + if cls._instance is None: + cls._instance.__call__(*args, **kwargs) + return cls._instance Afterwards, this file is obsolete. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#48 (review)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAt-cdUVF7uFS_0tZTWiOOVnzxtPhBp9ks5vc9cagaJpZM4cUuQK> .

[kernc]

No, no, I love how few extra lines are needed to define those classes! Just test pdoc API (pdoc.Class.subclasses) instead of Python API (type dunder methods). 😄

[kernc]

Sorry, with any_module I meant it can be a dummy module just so long the constructor executes.

[kernc]

Applied with slight modification. I see I have been oversimplifying my examples. Thanks for figuring it out! 😄