Should implicit converting an instance of int, float, complex, str, bytes, etc subclasses to call appropriate special method __int__ (or __index__), __float__, __complex__, __str__, __bytes__, etc? Currently explicit converting calls these methods, but implicit converting doesn't.

class I(int):
...     def __int__(self): return 42
...     def __index__(self): return 43
...
class F(float):
...     def __float__(self): return 42.0
...
class S(str):
...     def __str__(self): return '*'
...
int(I(65))
42
float(F(65))
42.0
str(S('A'))
'*'
chr(I(65))
'A'
import cmath; cmath.rect(F(65), 0)
(65+0j)
ord(S('A'))
65

Issue17576 [1] proposes to call special methods for implicit converting. I have doubts about this.

1. I afraid that this will adds places where arbitrary Python code is unexpectedly called. For example see changeset9a61be172c23 discussed in neighbor thread. If the "k" format code will call __int__(), Python code can modify unpacked list argument during parsing arguments in PyArg_ParseTuple().

2. PyLong_As*() functions already is not very consistent. Some of them calls __int__() for argument which is not an instance of int subclass, other accepts only instances of int subclasses. PyLong_AsVoidPtr() calls or not calls __int__() depending on the sign of the argument.

3. We can't consistency call special methods for all types. E.g. for strings we can't call __str__() when processing the "s" code in PyArg_ParseTuple() because this will cause a leak.

I think that overriding special converting method in a subclass of corresponding type should be restricted. I see two consistent and safe possibilities:

1. Forbidden. I.e. above declarations of I, F and S classes should raise exceptions.

2. Has no effect. I.e. both int(I(65)) and operator.index(I(65)) should return 65.

[1] http://bugs.python.org/issue17576

_______________________________________________
Python-Dev mailing list
Python-Dev@python.org
https://mail.python.org/mailman/listinfo/python-dev
Unsubscribe: 
https://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com

Reply via email to