Please consider adding an overrideable flag to abstractmethod
Marking a method M declared in C with abstractmethod indicates that M needs to be *implemented* in any subclass D of C for D to be instantiated.
We usually think of overriding a method N to mean replacing one implementation in some class E with another in some subclass of E, F. Often, the subclass implementation calls super to add behavior rather than replace it.
I think that this concept of *implementing* is different than *overriding*.
However, abstract classes can have reasonable definition, and should sometimes be overridden in the sense that subclasses should call super. For example, when inheriting from AbstractContextManager, you need to *override* the abstractmethod (!) __exit__, and if you want your class to work polymorphically, you should call super.
This is extremely weird. Understandably, the pylint people are confused by it (https://github.com/PyCQA/pylint/issues/1594) and raise bad warnings.
It also makes it impossible for me to raise warnings in my ipromise (https://github.com/NeilGirdhar/ipromise) project. See, for example, https://github.com/NeilGirdhar/ipromise/blob/master/ipromise/test/test_overr... classes Y and W, which ought to raise, but that would raise on reasonable code.
My suggestion is to add a rarely used flag to abstractmethod:
class AbstractContextManager: @abstractmethod(overrideable=True) def __exit__(self, exc_type, exc_value, traceback): pass
This would set a flag on the method like __abstractmethod_overrideable__, which could be checked by ipromise's @overrides decorator, pylint's call check, and anyone else that wants to know that a method should be overridden.
Best,
Neil
Alternatively, the need for an overriding implementation to call super could be marked by a different decorator. This would allow linters to check that subclasses call super when they should, which is a very useful check in my opinion. Such a decorator could be called "@overrideable", and could be exposed in abc. When used in combination with @abstractmethod, it could have the above behavior.
On Sunday, February 17, 2019 at 5:28:49 AM UTC-5, Neil Girdhar wrote:
Marking a method M declared in C with abstractmethod indicates that M needs to be *implemented* in any subclass D of C for D to be instantiated.
We usually think of overriding a method N to mean replacing one implementation in some class E with another in some subclass of E, F. Often, the subclass implementation calls super to add behavior rather than replace it.
I think that this concept of *implementing* is different than *overriding*.
However, abstract classes can have reasonable definition, and should sometimes be overridden in the sense that subclasses should call super. For example, when inheriting from AbstractContextManager, you need to *override* the abstractmethod (!) __exit__, and if you want your class to work polymorphically, you should call super.
This is extremely weird. Understandably, the pylint people are confused by it (https://github.com/PyCQA/pylint/issues/1594) and raise bad warnings.
It also makes it impossible for me to raise warnings in my ipromise ( https://github.com/NeilGirdhar/ipromise) project. See, for example, https://github.com/NeilGirdhar/ipromise/blob/master/ipromise/test/test_overr... classes Y and W, which ought to raise, but that would raise on reasonable code.
My suggestion is to add a rarely used flag to abstractmethod:
class AbstractContextManager: @abstractmethod(overrideable=True) def __exit__(self, exc_type, exc_value, traceback): pass
This would set a flag on the method like __abstractmethod_overrideable__, which could be checked by ipromise's @overrides decorator, pylint's call check, and anyone else that wants to know that a method should be overridden.
Best,
Neil
participants (1)
-
Neil Girdhar