Safe Refactoring

Python’s type system does not provide a way to identify call sites that need to be changed to stay consistent when an overridden function API changes. This makes refactoring and transforming code more dangerous.

Consider this simple inheritance structure:

class Parent:
    def foo(self, x: int) -> int:
        return x

class Child(Parent):
    def foo(self, x: int) -> int:
        return x + 1

def parent_callsite(parent: Parent) -> None:
    parent.foo(1)

def child_callsite(child: Child) -> None:
    child.foo(1)

If the overridden method on the superclass is renamed or deleted, type checkers will only alert us to update call sites that deal with the base type directly. But the type checker can only see the new code, not the change we made, so it has no way of knowing that we probably also needed to rename the same method on child classes.

A type checker will happily accept this code, even though we are likely introducing bugs:

class Parent:
    # Rename this method
    def new_foo(self, x: int) -> int:
        return x

class Child(Parent):
    # This (unchanged) method used to override `foo` but is unrelated to `new_foo`
    def foo(self, x: int) -> int:
        return x + 1

def parent_callsite(parent: Parent) -> None:
    # If we pass a Child instance we’ll now run Parent.new_foo - likely a bug
    parent.new_foo(1)

def child_callsite(child: Child) -> None:
    # We probably wanted to invoke new_foo here. Instead, we forked the method
    child.foo(1)

This code will type check, but there are two potential sources of bugs:

• If we pass a Child instance to the parent_callsite function, it will invoke the implementation in Parent.new_foo. rather than Child.foo. This is probably a bug - we presumably would not have written Child.foo in the first place if we didn’t need custom behavior.
• Our system was likely relying on Child.foo behaving in a similar way to Parent.foo. But unless we catch this early, we have now forked the methods, and in future refactors it is likely no one will realize that major changes to the behavior of new_foo likely require updating Child.foo as well, which could lead to major bugs later.

The incorrectly-refactored code is type-safe, but is probably not what we intended and could cause our system to behave incorrectly. The bug can be difficult to track down because our new code likely does execute without throwing exceptions. Tests are less likely to catch the problem, and silent errors can take longer to track down in production.

We are aware of several production outages in multiple typed codebases caused by such incorrect refactors. This is our primary motivation for adding an @override decorator to the type system, which lets developers express the relationship between Parent.foo and Child.foo so that type checkers can detect the problem.

Subclass Implementations Become More Explicit

We believe that explicit overrides will make unfamiliar code easier to read than implicit overrides. A developer reading the implementation of a subclass that uses @override can immediately see which methods are overriding functionality in some base class; without this decorator, the only way to quickly find out is using a static analysis tool.

Precedent in Other Languages and Runtime Libraries

Disadvantages

Using @override will make code more verbose.

No New Rules for Override Compatibility

This PEP is exclusively concerned with the handling of the new @override decorator, which specifies that the decorated method must override some attribute in an ancestor class. This PEP does not propose any new rules regarding the type signatures of such methods.

Motivation

The primary reason for a strict mode that requires @override is that developers can only trust that refactors are override-safe if they know that the @override decorator is used throughout the project.

There is another class of bug related to overrides that we can only catch using a strict mode.

Consider the following code:

class Parent:
    pass

class Child(Parent):
    def foo(self) -> int:
        return 2

Imagine we refactor it as follows:

class Parent:
    def foo(self) -> int:   # This method is new
        return 1

class Child(Parent):
    def foo(self) -> int:  # This is now an override!
        return 2

def call_foo(parent: Parent) -> int:
    return parent.foo()  # This could invoke Child.foo, which may be surprising.

The semantics of our code changed here, which could cause two problems:

• If the author of the code change did not know that Child.foo already existed (which is very possible in a large codebase), they might be surprised to see that call_foo does not always invoke Parent.foo.
• If the codebase authors tried to manually apply @override everywhere when writing overrides in subclasses, they are likely to miss the fact that Child.foo needs it here.

At first glance this kind of change may seem unlikely, but it can actually happen often if one or more subclasses have functionality that developers later realize belongs in the base class.

With a strict mode, we will always alert developers when this occurs.

Precedent

Most of the typed, object-oriented programming languages we looked at have an easy way to require explicit overrides throughout a project:

• C#, Kotlin, Scala, and Swift always require explicit overrides
• TypeScript has a –no-implicit-override flag to force explicit overrides
• In Hack and Java the type checker always treats overrides as opt-in, but widely-used linters can warn if explicit overrides are missing.

Set __override__ = True when possible

At runtime, @typing.override will make a best-effort attempt to add an attribute __override__ with value True to its argument. By “best-effort” we mean that we will try adding the attribute, but if that fails (for example because the input is a descriptor type with fixed slots) we will silently return the argument as-is.

This is exactly what the @typing.final decorator does, and the motivation is similar: it gives runtime libraries the ability to use @override. As a concrete example, a runtime library could check __override__ in order to automatically populate the __doc__ attribute of child class methods using the parent method docstring.

Limitations of setting __override__

As described above, adding __override__ may fail at runtime, in which case we will simply return the argument as-is.

In addition, even in cases where it does work, it may be difficult for users to correctly work with multiple decorators, because successfully ensuring the __override__ attribute is set on the final output requires understanding the implementation of each decorator:

• The @override decorator needs to execute after ordinary decorators like @functools.lru_cache that use wrapper functions, since we want to set __override__ on the outermost wrapper. This means it needs to go above all these other decorators.
• But @override needs to execute before many special descriptor-based decorators like @property, @staticmethod, and @classmethod.
• As discussed above, in some cases (for example a descriptor with fixed slots or a descriptor that also wraps) it may be impossible to set the __override__ attribute at all.

As a result, runtime support for setting __override__ is best effort only, and we do not expect type checkers to validate the ordering of decorators.

Rely on Integrated Development Environments for safety

Modern Integrated Development Environments (IDEs) often provide the ability to automatically update subclasses when renaming a method. But we view this as insufficient for several reasons:

• If a codebase is split into multiple projects, an IDE will not help and the bug appears when upgrading dependencies. Type checkers are a fast way to catch breaking changes in dependencies.
• Not all developers use such IDEs. And library maintainers, even if they do use an IDE, should not need to assume pull request authors use the same IDE. We prefer being able to detect problems in continuous integration without assuming anything about developers’ choice of editor.

Runtime enforcement

We considered having @typing.override enforce override safety at runtime, similarly to how @overrides.overrides does today.

We rejected this for four reasons:

• For users of static type checking, it is not clear this brings any benefits.
• There would be at least some performance overhead, leading to projects importing slower with runtime enforcement. We estimate the @overrides.overrides implementation takes around 100 microseconds, which is fast but could still add up to a second or more of extra initialization time in million-plus line codebases, which is exactly where we think @typing.override will be most useful.
• An implementation may have edge cases where it doesn’t work well (we heard from a maintainer of one such closed-source library that this has been a problem). We expect static enforcement to be simple and reliable.
• The implementation approaches we know of are not simple. The decorator executes before the class is finished evaluating, so the options we know of are either to inspect the bytecode of the caller (as @overrides.overrides does) or to use a metaclass-based approach. Neither approach seems ideal.

Mark a base class to force explicit overrides on subclasses

We considered including a class decorator @require_explicit_overrides, which would have provided a way for base classes to declare that all subclasses must use the @override decorator on method overrides. The Overrides library has a mixin class, EnforceExplicitOverrides, which provides similar behavior in runtime checks.

We decided against this because we expect owners of large codebases will benefit most from @override, and for these use cases having a strict mode where explicit @override is required (see the Backward Compatibility section) provides more benefits than a way to mark base classes.

Moreover we believe that authors of projects who do not consider the extra type safety to be worth the additional boilerplate of using @override should not be forced to do so. Having an optional strict mode puts the decision in the hands of project owners, whereas the use of @require_explicit_overrides in libraries would force project owners to use @override even if they prefer not to.

Include the name of the ancestor class being overridden

We considered allowing the caller of @override to specify a specific ancestor class where the overridden method should be defined:

class Parent0:
    def foo(self) -> int:
        return 1


class Parent1:
    def bar(self) -> int:
        return 1


class Child(Parent0, Parent1):
    @override(Parent0)  # okay, Parent0 defines foo
    def foo(self) -> int:
        return 2

    @override(Parent0)  # type error, Parent0 does not define bar
    def bar(self) -> int:
        return 2

This could be useful for code readability because it makes the override structure more explicit for deep inheritance trees. It also might catch bugs by prompting developers to check that the implementation of an override still makes sense whenever a method being overridden moves from one base class to another.

We decided against it because:

• Supporting this would add complexity to the implementation of both @override and type checker support for it, so there would need to be considerable benefits.
• We believe that it would be rarely used and catch relatively few bugs.
  • The author of the Overrides package has noted that early versions of his library included this capability but it was rarely useful and seemed to have little benefit. After it was removed, the ability was never requested by users.