Problem subclassing tuple

[Terry Reedy]

"John Wilson" <tug at wilson.co.uk> wrote in message
news:mailman.1051631814.18940.python-list at python.org...
> Given this class:
>
> class Holder(tuple):
>     def __init__(self, *values):
>         tuple.__init__(self, values)
>
>
> When I try to create an instance:
>
> Holder(1, 2)
>
> I get:
>
> TypeError: tuple() takes at most 1 argument (2 given)
>
> It looks like the constructor for the superclass is being called
rather than
> the __init__ method on Holder.
>
> I'm obviously missing something here. Do I have to supply a __new__
method?
> Any help would be appreciated.


Because tuples are immutable, one cannot mutate them in an init
method, even to set the initial values.  So, as you apparently
suspected, they are initialized in the __new__ method from a sequence.
This is the source of the exception (see below). tuple.__init__ should
be a no op.

>>> class Holder(tuple):
...      def __init__(self, *values):
...          print "Holder init entered"
...
>>> Holder(1,2)
Traceback (most recent call last):
  File "<stdin>", line 1, in ?
TypeError: tuple() takes at most 1 argument (2 given)
# init not entered
>>> Holder((1,2))
Holder init entered
(1, 2)

While Holder.__init__() cannot modify the value of self, it *can* add
attributes (which surprised me a bit) so it can have a use.

>>> class Holder(tuple):
...     def __init__(self, seq):
...         self.y = 1
...
>>> a=Holder((1,2))
>>> a.y
1

I don't know what more you could do in a __new__ method.

Terry J. Reedy