Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Document we're not tracking relationships between symbols #16018

Merged
Merged
Changes from 1 commit
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Next Next commit
Document we're not tracking relationships between symbols
  • Loading branch information
ikonst committed Sep 2, 2023
commit d5e9302f5b7d2238d99ce17c82846a76b62b7966
43 changes: 38 additions & 5 deletions docs/source/type_narrowing.rst
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,7 @@
Type narrowing
==============

This section is dedicated to several type narrowing
This section is dedicated to several type narrowing
techniques which are supported by mypy.

Type narrowing is when you convince a type checker that a broader type is actually more specific, for instance, that an object of type ``Shape`` is actually of the narrower type ``Square``.
Expand All @@ -14,10 +14,11 @@ Type narrowing expressions

The simplest way to narrow a type is to use one of the supported expressions:

- :py:func:`isinstance` like in ``isinstance(obj, float)`` will narrow ``obj`` to have ``float`` type
- :py:func:`issubclass` like in ``issubclass(cls, MyClass)`` will narrow ``cls`` to be ``Type[MyClass]``
- :py:class:`type` like in ``type(obj) is int`` will narrow ``obj`` to have ``int`` type
- :py:func:`callable` like in ``callable(obj)`` will narrow object to callable type
- :py:func:`isinstance` like in :code:`isinstance(obj, float)` will narrow ``obj`` to have ``float`` type
- :py:func:`issubclass` like in :code:`issubclass(cls, MyClass)` will narrow ``cls`` to be ``Type[MyClass]``
- :py:class:`type` like in :code:`type(obj) is int` will narrow ``obj`` to have ``int`` type
- :py:func:`callable` like in :code:`callable(obj)` will narrow object to callable type
- :code:`obj is not None` will narrow object to its :ref:`non-optional form <strict_optional>`
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

strangely we didn't mention this


Type narrowing is contextual. For example, based on the condition, mypy will narrow an expression only within an ``if`` branch:

Expand Down Expand Up @@ -83,6 +84,38 @@ We can also use ``assert`` to narrow types in the same context:
reveal_type(x) # Revealed type is "builtins.int"
print(x + '!') # Typechecks with `mypy`, but fails in runtime.


Limitations of narrowing
~~~~~~~~~~~~~~~~~~~~~~~~
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hope this is the right place. I wanted it to have a heading, to make it easy to refer people, but it's also a bit strange that it comes as a topic in the same hierarchy and right before "is_subclass" and "callable".

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Let's make it a heading and move it to the bottom of the page.


Mypy's analysis is limited to individual symbols and it will not track
relationships between symbols. For example, in the following code
it's easy to deduce that if :code:`a` is None then :code:`b` must not be,
therefore :code:`a or b` will always be a string, but Mypy will not be able to tell that:

.. code-block:: python

def f(a: str | None, b: str | None) -> str:
if a is not None or b is not None:
return a or b # Incompatible return value type (got "str | None", expected "str")
return 'spam'

Tracking these sort of cross-variable conditions in a type checker would add significant complexity
and performance overhead and would be computationally infeasible in all but the most basic cases.
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
and performance overhead and would be computationally infeasible in all but the most basic cases.
and performance overhead.

Claim feels a little too strong, first half of the sentence says enough.


You may override the type checker with a :ref:`cast <casts>`, or rewrite the function to be
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
You may override the type checker with a :ref:`cast <casts>`, or rewrite the function to be
You may override the type checker with an ``assert`` or :ref:`cast <casts>`, or rewrite the function to be

When I run into this, assert is usually more ergonomic. It's also safer than cast.

Copy link
Contributor Author

@ikonst ikonst Sep 2, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Probably would phrase as "help the type checker with an assert", but I'm not sure it'll be more ergonomic in this case. Maybe I'm tired and missing something obvious?

cast:

def f(a: str | None, b: str | None) -> str:
    if a is not None or b is not None:
        return a or cast(str, b)
    return 'spam'

assert:

def f(a: str | None, b: str | None) -> str:
    if a is not None or b is not None:
        if a:
            return a
        assert b is not None
        return b
    return 'spam'

third option:

def f(a: str | None, b: str | None) -> str:
    if a is not None or b is not None:
        return a or b or 'impossible'
    return 'spam'

Copy link
Collaborator

@hauntsaninja hauntsaninja Sep 2, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah, not more ergonomic in this case, but better in the cases where I usually run into this. But maybe I dislike cast more than most people :-)

I'd do assert a or b; return a or b

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

assert a or b doesn't help you either because we don't narrow conditionally. (Also, in this case you'd get more than you'd bargain for, as empty b would assert.)

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Oh oops, I guess you'd have to do ret = a or b; assert ret is not None; return ret

slightly more verbose:

.. code-block:: python

def f(a: str | None, b: str | None) -> str:
if a is not None:
return a
elif b is not None:
return b
return 'spam'


issubclass
~~~~~~~~~~

Expand Down