This is not as easy as it seems, some core library functions don't work when only str is overwritten (checked with Python 2.7), see this thread for examples How to make a class JSON serializable Also, try this
import json
class A(unicode):
def __str__(self):
return 'a'
def __unicode__(self):
return u'a'
def __repr__(self):
return 'a'
a = A()
json.dumps(a)
produces
'""'
and not
'"a"'
as would be expected.
EDIT: answering mchicago's comment:
unicode does not have any attributes -- it is an immutable string, the value of which is hidden and not available from high-level Python code. The json module uses re for generating the string representation which seems to have access to this internal attribute. Here's a simple example to justify this:
b = A('b')
print b
produces
'a'
while
json.dumps({'b': b})
produces
{"b": "b"}
so you see that the internal representation is used by some native libraries, probably for performance reasons.
See also this for more details: http://www.laurentluce.com/posts/python-string-objects-implementation/