Diez B. Roggisch wrote:
> Ron Garret wrote:
>> Is this a bug or a feature?
>>
>> class mydict(dict):
>> def __setitem__(self, key, val):
>> print 'foo'
>> dict.__setitem__(self, key, val)
>>
>>
>>>>>d=mydict()
>>>>>d[1]=2
>>
>> foo
>>
>>>>>d.setdefault(2,3)
>
>
> Feature. If it wouldn't bypass __setitem__, how exactly would you make a
> default-item? Using __setitem__ implies a key. So if setdefault
> was implemented as
>
> def setdefault(self, v):
> self["SOME_DEFAULT_KEY_NAME"] = v
>
> and later on one writes e.g. a HTML-page with a form input field named
> "SOME_DEFAULT_KEY_NAME" that gets stored in a dict - it would overwrite
> the default value.
>
> So it has to bypass __setitem__, as otherwise it can't distinguish
> between "real" and the default value - the latter one is not allowed to
> have a key that is in any imaginable way used by the user.
The implementation is certainly a design decision. setdefault() could be
implemented in terms of __set/getitem__() as
def setdefault(self, key, value=None):
try:
return self[key]
except KeyError:
self[key] = value
return self[key]
I guess it's not done for performance reasons.
Peter
--
http://mail.python.org/mailman/listinfo/python-list
