Python: Extending a predefined named tuple

It's quite easy to knock something together that allows you to compose namedtuples from other namedtuples as well as introduce new fields.

def extended_namedtuple(name, source_fields):
    assert isinstance(source_fields, list)
    new_type_fields = []
    for f in source_fields:
        try:
            new_type_fields.extend(f._fields)
        except:
            new_type_fields.append(f) 
    return namedtuple(name, new_type_fields) 

# source types
Name = namedtuple('Name', ['first_name', 'last_name'])
Address = namedtuple('Address', ['address_line1', 'city'])
# new type uses source types and adds additional ID field
Customer = extended_namedtuple('Customer', ['ID', Name, Address])
# using the new type
cust1 = Customer(1, 'Banana', 'Man', '29 Acacia Road', 'Nuttytown')
print(cust1)

This outputs the following :

Customer(ID=1, first_name='Banana', last_name='Man', address_line1='29 Acacia Road', city='Nuttytown')

You can subclass a namedtuple-produced class, but you need to study the generated class more closely. You'll need to add another __slots__ attribute with the extra fields, update the _fields attribute, create new __repr__ and _replace methods (they hardcode the field list and class name) and add extra property objects for the additional fields. See the example in the documentation.

That's all a little too much work. Rather than subclass, I'd just reuse the somenamedtuple._fields attribute of the source type:

LookupElement = namedtuple('LookupElement', ReadElement._fields + ('lookups',))

The field_names argument to the namedtuple() constructor doesn't have to be a string, it can also be a sequence of strings. Simply take the _fields and add more elements by concatenating a new tuple.

Demo:

>>> from collections import namedtuple
>>> ReadElement = namedtuple('ReadElement', 'address value')
>>> LookupElement = namedtuple('LookupElement', ReadElement._fields + ('lookups',))
>>> LookupElement._fields
('address', 'value', 'lookups')
>>> LookupElement('addr', 'val', 'lookup') 
LookupElement(address='addr', value='val', lookups='lookup')

This does mean that the extended type is not a subclass of the base type. If you must have a class hierarchy, then rather than try to make named tuples fit that model, I'd switch to using dataclasses instead. Dataclasses can serve the same purpose in most usecases named tuples are used for, but can easily be subclassed.

Extending Martijn Pieters' answer: there is a way to make the new namedtuple class a subclass of the other, without having to hack. Simply create the new namedtuple separately, and then use its __new__ method instead of using super:

from collections import namedtuple

class ReadElement(namedtuple('ReadElement', ('address', 'value'))):
    def compute(self):
        return self.value + 1

_LookupElement = namedtuple('_LookupElement', ReadElement._fields + ('lookups',))

class LookupElement(_LookupElement, ReadElement):
    def __new__(self, address, value, lookups):
        return _LookupElement.__new__(LookupElement, address, value, lookups)

assert issubclass(LookupElement, ReadElement)
l = LookupElement('ad', 1, dict())
assert isinstance(l, ReadElement)
assert l.compute() == 2

It seems that this also works without even overriding __new__ !

from collections import namedtuple

class ReadElement(namedtuple('ReadElement', ('address', 'value'))):
    def compute(self):
        return self.value + 1

class LookupElement(namedtuple('LookupElement', ReadElement._fields + ('lookups',)),
                    ReadElement):
    """nothing special to do"""
    pass