mypy：为什么“int”是“float”的子类型？

Question

Why does "mypy" consider "int" as a subtype of "float"? A subtype shall support all methods of its supertype, but "float" has methods, which "int" does not support:

test.py:

def f(x : float) -> bool:
    return x.is_integer()

print(f(123.0))
print(f(123))

The static type checker accepts passing an "int" argument for a "float" parameter:

(3.8.1) myhost% mypy test.py
Success: no issues found in 1 source file

But this does not guarantee, that there are no errors at runtime:

(3.8.1) myhost% python test.py
True
Traceback (most recent call last):
  File "test.py", line 5, in <module>
    print(f(123))
  File "test.py", line 2, in f
    return x.is_integer()
AttributeError: 'int' object has no attribute 'is_integer'

because "float" has additional methods, which "int" does not have.

Answer 1

'Why does "mypy" consider "int" as a subtype of "float"?'

Because practicality has so far been considered to beat purity here. This is not to say that one could not propose that typing define a Scalar type that would include ints and floats but only be valid for arithmetic operations.

Note that int / int was changed in 3.0 so that float(int / int) == float(int) / float(int), to make int and float arithmetic consistent for equal int and float values.

Note also that a type-check passing does not mean no runtime errors: division by zero and overflow are still possible, as well as many others.

Answer 2

As @juanpa.arrivillaga pointed out, the explanation is on https://mypy.readthedocs.io/en/latest/duck_type_compatibility.html .

A subtype shall support all methods of its supertype, but "float" has methods, which "int" does not support

int is not a subtype of float , so it doesn't have to support methods of float .

The mechanism is good because passing integer values shouldn't cause errors, unless you really want them as in your example. You explicitly tried to use a method which doesn't exist. In common situations, we only make arithmetic operations on numbers, so a problem rarely exists and you can always avoid it by adding .0 as you wrote.

It is a common behavior in most languages to assume that int is a special case of float , consider for example C++ int to float implicit conversion.