Issue 26983: float() can return not exact float instance

The float constructor can return an instance of float subclass.

>>> class FloatSubclass(float):
...     pass
... 
>>> class BadFloat:
...     def __float__(self):
...         return FloatSubclass(1.2)
... 
>>> type(float(BadFloat()))
<class '__main__.FloatSubclass'>

Comparing with other types, complex() always returns complex:

>>> class ComplexSubclass(complex):
...     pass
... 
>>> class BadComplex:
...     def __complex__(self):
...         return ComplexSubclass(1.2, 3.4)
... 
>>> type(complex(BadComplex()))
<class 'complex'>

And int() can return an instance of int subclass, but this behavior is deprecated:

>>> class BadInt:
...     def __int__(self):
...         return True
... 
>>> int(BadInt())
__main__:1: DeprecationWarning: __int__ returned non-int (type bool).  The ability to return an instance of a strict subclass of int is deprecated, and may be removed in a future version of Python.
True

May be we should either deprecate __float__ returning non-float (as for int), or convert the result to exact float (as for complex).

The constructor of float subclass always returns an instance of correct type.

>>> class FloatSubclass2(float):
...     pass
... 
>>> type(FloatSubclass2(BadFloat()))
<class '__main__.FloatSubclass2'>

Proposed patch makes float() always returning exact float.

+1 for the idea, and the patch LGTM.

I was initially a bit confused by the wording of the warning: presumably, we're not going to change Python to make returning an instance of a strict float subclass from __float__ illegal (I don't really see how that would be possible); we're just going to check the return type at the internal call sites to __float__ and raise if we get something other than an exact float. That is, I'd still expect this code to work on future versions of Python:

>>> class MyFloat(float): pass
... 
>>> class A:
...     def __float__(self): return MyFloat()
... 
>>> a = A()
>>> a.__float__()
0.0

But these would both become an error in Python 3.7 (say):

>>> float(a)
0.0
>>> math.sqrt(a)
0.0

Does that match your thinking?

We should probably add issues for checking and fixing other places that __float__ is used internally (like the math module).

The warning message is copied from __int__. As I understand, the conclusion of the Python-Dev discussion is that __int__, __float__, __complex__, etc should return exact int, float, complex, not subclasses. I have another patch that adds a warning for __complex__, and I am working on patches for other special methods. If my understanding is wrong, we should remove the deprecation warning for __int__.

> As I understand, the conclusion of the Python-Dev discussion is that __int__, __float__, __complex__, etc should return exact int, float, complex, not subclasses.

Agreed; that matches my understanding. I'm only observing that it would be difficult (and likely undesirable) to actually enforce that `__int__` itself always returns an exact int: I'm not sure where that check would/could take place, and I'd expect that a direct user-level call to `my_object.__int__` probably wouldn't be subject to such a check. What we *can* do is check that the result of the `nb_int` call is always an int in the places we use it.

Or we could even go further, and write custom slot_nb_float and slot_nb_int functions in Objects/typeobject.c that incorporates those checks. What I don't think we can do is check that a *direct* call to `__int__` or `__float__` always returns something of the exact type.

Here is updated patch. Differences:

* Deprecation warning is now emitted in functions accepting floats (like math.sqrt).
* Improved error messages. Now reported wrong type name and a type of wrong __float__ method.
* float() now returns the same object for exact float argument.

Could you propose better warning message?

Can you please attach the new patch?

Oh, sorry, how could I miss it?

Ping.

float_exact2.patch LGTM

New changeset 050e5f803999 by Serhiy Storchaka in branch 'default':
Issue #26983: float() now always return an instance of exact float.
https://hg.python.org/cpython/rev/050e5f803999

Thank you for your review Mark.

test_format resulted in semi-failure due to this change. The attached patch fixes the issue.

New changeset 6216fb8afa53 by Serhiy Storchaka in branch 'default':
Issue #26983: Fixed test_format failure.
https://hg.python.org/cpython/rev/6216fb8afa53

My failure. Thank you for your report and patch SilentGhost.

assignee: serhiy.storchaka

nosy: + SilentGhost
messages: + msg267521
resolution: fixed -> (no value)
stage: resolved -> commit review

messages: + msg265318

stage: patch review