How to use spec when mocking data classes in Python

Question

I'm trying to port our namedtuple classes into dataclass in Python 3.6 using the backport package. However, I noticed when mocking dataclass classes, you cannot use the "spec" keyword anymore. I assume it's because the dataclass code is auto generated.

from dataclasses import dataclass
import mock

@dataclass
class A:
    aaa: str
    bbb: int


m = mock.Mock(spec=A)

m.aaa

And this is the error I get:

AttributeError: Mock object has no attribute 'aaa'

Any idea if there's any way to automatically set all the attributes from original object to the mock object? I have lots of data classes with a lot of data. It's going to be really tedious if I try to manually set the values one by one.

Mollymollycoddle

That wouldn't have worked anyway even without dataclass. – Anlage

If it were a regular class, it would have returned something like this, when trying to access m.aaa: <Mock name='mock.aaa' id='139990326478496'> – Mollymollycoddle

Oh, you were trying to access instance variables. In my case I was using class variables, similar to what namedtuple and dataclasse classes are defined. – Mollymollycoddle

Class variables also should work pretty much the same with or without dataclass. You just don't have any class variables here. (aaa: str is an annotation, not a class variable, and the dataclass processing doesn't create an aaa class variable.) – Anlage

Thank you for providing the clarification. – Mollymollycoddle

Answer 1

I ended up using this generic helper function to achieve what spec does with regular classes:

import mock
from dataclasses import fields


def create_dataclass_mock(obj):
    return mock.Mock(spec=[field.name for field in fields(obj)])

Mollymollycoddle

This is no better than just using mock.Mock() directly with no spec. It won't prevent accesses like mock_obj.attribute_that_should_not_exist. – Anlage

I just tested this and it works! Thanks. Code snippet with a working example in python 3.7 – Anny

But if the actual (non-test) code uses fields(MyDataClass), this will not work. – Core

Update: to resolve this, I was able to patch dataclasses.fields with return_value=[MockField(name) for name in MyMock._spec_signature.parameters.keys()] where MockField = namedtuple('Field', ['name']). – Core

Answer 2

You can also pass an instance with dummy values to spec

from unittest.mock import Mock
from dataclasses import dataclass

@dataclass
class A:
    aaa: str
    bbb: int

m = Mock(spec=A(None, None))

print(m.bbb)
# <Mock name='mock.bbb' id='139766470904856'>

Elroyels

Answer 3

Based on the answer from mohi666. If you also want to prevent setting Mock attributes not specified in the dataclass, use spec_set instead of spec:

import mock
from dataclasses import dataclass, fields

@dataclass
class A:
   x: str


def create_dataclass_mock(obj):
    return mock.Mock(spec_set=[field.name for field in fields(obj)])

m = create_dataclass_mock(A)
m.x = 'test' # works
m.y = 'another test' # raises AttributeError

Popele

Answer 4

Dataclass fields are actually implemented as instance variables. You have 4 options:

• give the field a default value:

@dataclass()
class MyDataClass1:
    my_field: str = field(default="my_field_val")
m = Mock(spec=MyDataClass1("my_field_val"))
print(m.my_field)

• provide the value for the field as an arg in the constructor call of the dataclass:

@dataclass()
class MyDataClass2:
    my_field: str = field()
m = Mock(spec=MyDataClass2("my_field_val"))
print(m.my_field)

• Instantiate the mock:

@dataclass()
class MyDataClass3:
    my_field: str = field()
m = Mock(spec=MyDataClass3)
m_inst = m()
print(m_inst.my_field)

• Use a list comprehension to instantiate all fields in the spec:

@dataclass()
class MyDataClass4:
    my_field: str = field()
m = Mock(spec_set=[field.name for field in fields(MyDataClass4)])
print(m.my_field)

Core