向构造函数传递太多参数是否被视为反模式？

Question

I am considering using the factory_boy library for API testing. An example from the documentation is:

class UserFactory(factory.Factory):
    class Meta:
        model = base.User

    first_name = "John"
    last_name = "Doe"

For this to work, we need first_name , last_name , etc to be passed as parameters to the __init__() method of the base.User() class . However, if you have many parameters this leads to something like:

class User(object):

    GENDER_MALE = 'mr'
    GENDER_FEMALE = 'ms'

    def __init__(self, title=None, first_name=None, last_name=None, is_guest=None,
             company_name=None, mobile=None, landline=None, email=None, password=None,
             fax=None, wants_sms_notification=None, wants_email_notification=None,
             wants_newsletter=None, street_address=None):

        self. title = title
        self.first_name = first_name
        self.last_name = last_name
        self.company_name = company_name
        self.mobile = mobile
        self.landline = landline
        self.email = email
        self.password = password
        self.fax = fax
        self.is_guest = is_guest
        self.wants_sms_notification = wants_sms_notification
        self.wants_email_notification = wants_email_notification
        self.wants_newsletter = wants_newsletter
        self.company_name = company_name
        self.street_address = street_address

Now the question is, is this construction considered anti-pattern, and if yes, what alternatives do I have?

Thanks

Answer 1

In Python 3.7, dataclasses (specified in PEP557 ) were added. This allows you to only write these arguments once and not again in the constructor, since the constructor is made for you:

from dataclasses import dataclass

@dataclass
class User:
    title: str = None
    first_name: str = None
    last_name: str = None
    company_name: str = None
    mobile: str = None
    landline: str = None
    email: str = None
    password: str = None
    fax: str = None
    is_guest: bool = True
    wants_sms_notification: bool = False
    wants_email_notification: bool = False
    wants_newsletter: bool = False
    street_address: str = None

It also adds a __repr__ to the class as well as some others. Note that explicitly inheriting from object is no longer needed in Python 3, since all classes are new-style classes by default.

There are a few drawbacks, though. It is slightly slower on class definition (since these methods need to be generated). You need to either set a default value or add a type annotation , otherwise you get a name error. If you want to use a mutable object, like a list, as a default argument, you need to use dataclass.field(default_factory=list) (normally it is discouraged to write eg def f(x=[]) , but here it actually raises an exception).

This is useful where you have to have all those arguments in the constructor, because they all belong to the same object and cannot be extracted to sub-objects, for example.

Answer 2

You could pack the __init__ method's keyword arguments into one dict, and set them dynamically with setattr :

class User(object):
    GENDER_MALE = 'mr'
    GENDER_FEMALE = 'ms'
    def __init__(self, **kwargs):
        valid_keys = ["title", "first_name", "last_name", "is_guest", "company_name", "mobile", "landline", "email", "password", "fax", "wants_sms_notification", "wants_email_notification", "wants_newsletter","street_address"]
        for key in valid_keys:
            setattr(self, key, kwargs.get(key))

x = User(first_name="Kevin", password="hunter2")
print(x.first_name, x.password, x.mobile)

However, this has the drawback of disallowing you from supplying arguments without naming them - x = User("Mr", "Kevin") works with your original code, but not with this code.

Answer 3

Yes, too many arguments is an antipattern (as stated in Clean Code by RObert C. Martin)

To avoid this, you have two design approaches:

The essence pattern

The fluent interface/builder pattern

These are both similar in intent, in that we slowly build up an intermediate object, and then create our target object in a single step.

I'd recommend the builder pattern, it makes the code easy to read.

Answer 4

The biggest risk would be if you had a large number of positional arguments and then ended up not knowing which is which.. Keyword arguments definitely make this better.

As suggested by others, the builder pattern also works quite nicely. If you have a very large number of fields, you can also do something more generic, like so:

class Builder(object):

    def __init__(self, cls):
        self.attrs = {}
        self.cls = cls

    def __getattr__(self, name):
        if name[0:3] == 'set':
            def setter(x):
                field_name = name[3].lower() + name[4:]
                self.attrs[field_name] = x
                return self
            return setter
        else:
            return super(UserBuilder, self).__getattribute__(name)

    def build(self):
        return self.cls(**self.attrs)

class User(object):

    def __str__(self):
        return "%s %s" % (self.firstName, self.lastName)

    def __init__(self, **kwargs):
        # TODO: validate fields
        for key in kwargs:
            setattr(self, key, kwargs[key])

    @classmethod
    def builder(cls):
        return Builder(cls)

print (User.builder()
  .setFirstName('John')
  .setLastName('Doe')
  .build()) # prints John Doe

Answer 5

If overloading were not a problem then every class in python could be reduced to a single method, which we could call doIt (....). As with everything, it's best to do things in moderation. Overloading any method with umpteen arguments is bad practice. Instead, allow the user to build up the object in bite-size chunks of related data. It's more logical. In your case, you could split the calls into names, communications, and perhaps other.