问题:

class A():
   def __init__(self, data=''):
       self.data = data  

   def __str__(self):
       return str(self.data)

d = {}  
elem = A()  
d[elem] = 'abc'  

elem2 = A()
print d[elem2]    # KeyError  
# actually elem2! was used not elem

how can I implement this without error?

EDIT:
FFFUUU, the error was:
I tried to get d[elem2] (not elem) with another instance of A() BUT with the same content. (shame on me)
Still.. how can I do this? redefine __hash__?

回答1:

The answer is yes, you need to redefine __hash__():

>>> class A(object):
...   def __init__(self, data=''):
...     self.data = data
...   def __eq__(self, another):
...     return hasattr(another, 'data') and self.data == another.data
...   def __hash__(self):
...     return hash(self.data)
... 
>>> a1, a2, a3 = A('foo'), A('foo'), A('bar')
>>> d = {a1: 'foo'}
>>> d[a1]
'foo'
>>> d[a2]
'foo'
>>> d[a3]
Traceback (most recent call last):
  File "", line 1, in 
KeyError: __main__.A object at 0x927d0>

As explained in another comment default implementation of __hash__ is just simple identity, so if you want to make it more sophisticated, you need to define it explicitly.

回答2:

What you did should work, as long as you don't override the __hash__() and __eq__() methods. It will use object identity as equality. If you want a different notion of equality, you can override the __hash__() and __eq__() methods of your class.