Quick and dirty mock testing with mock_calls#

I needed to write a test that checked for a really complex sequence of mock calls for s3-credentials#3.

I ended up using the following trick, using pytest-mock:

1
def test_create(mocker):
2
    boto3 = mocker.patch("boto3.client")
3
    runner = CliRunner()
4
    with runner.isolated_filesystem():
5
        result = runner.invoke(cli, ["create", "pytest-bucket-simonw-1", "-c"])
6
        assert [str(c) for c in boto3.mock_calls] == [
7
            "call('s3')",
8
            "call('iam')",
9
            "call().head_bucket(Bucket='pytest-bucket-simonw-1')",
10
            "call().get_user(UserName='s3.read-write.pytest-bucket-simonw-1')",
11
            'call().put_user_policy(PolicyDocument=\'{"Version": "2012-10-17", "Statement": [{"Sid": "ListObjectsInBucket", "Effect": "Allow", "Action": ["s3:ListBucket"], "Resource": ["arn:aws:s3:::pytest-bucket-simonw-1"]}, {"Sid": "AllObjectActions", "Effect": "Allow", "Action": "s3:*Object", "Resource": ["arn:aws:s3:::pytest-bucket-simonw-1/*"]}]}\', PolicyName=\'s3.read-write.pytest-bucket-simonw-1\', UserName=\'s3.read-write.pytest-bucket-simonw-1\')',
12
            "call().create_access_key(UserName='s3.read-write.pytest-bucket-simonw-1')",
13
            "call().create_access_key().__getitem__('AccessKey')",
14
            "call().create_access_key().__getitem__().__str__()",
15
        ]

I used the trick I describe in How to cheat at unit tests with pytest and Black where I run that comparison against an empty [] list, then use pytest --pdb to drop into a debugger and copy and paste the output of [str(c) for c in boto3.mock_calls] into my test code.

Initially I used a comparison directly against boto3.mock_calls - but this threw a surprising error. The calls sequence I baked into my tests looked like this:

1
from unittest.mock import call
2

3
# ...
4

5
        assert boto3.mock_calls == [
6
            call("s3"),
7
            call("iam"),
8
            call().head_bucket(Bucket="pytest-bucket-simonw-1"),
9
            call().get_user(UserName="s3.read-write.pytest-bucket-simonw-1"),
10
            call().put_user_policy(
11
                PolicyDocument='{"Version": "2012-10-17", "Statement": [{"Sid": "ListObjectsInBucket", "Effect": "Allow", "Action": ["s3:ListBucket"], "Resource": ["arn:aws:s3:::pytest-bucket-simonw-1"]}, {"Sid": "AllObjectActions", "Effect": "Allow", "Action": "s3:*Object", "Resource": ["arn:aws:s3:::pytest-bucket-simonw-1/*"]}]}',
12
                PolicyName="s3.read-write.pytest-bucket-simonw-1",
13
                UserName="s3.read-write.pytest-bucket-simonw-1",
14
            ),
15
            call().create_access_key(UserName="s3.read-write.pytest-bucket-simonw-1"),
16
            call().create_access_key().__getitem__("AccessKey"),
17
            call().create_access_key().__getitem__().__str__(),
18
        ]

But when I ran pytest that last one failed:

1
E             -  'call().create_access_key().__getitem__()',
2
E             ?  -                                        ^
3
E             +  call().create_access_key().__getitem__().__str__(),
4
E             ?                                          ^^^^^^^^^^

It turns out __str__() calls do not play well with the call() constructor - see this StackOverflow question.

My solution was to cast them all to str() using a list comprehension, which ended up fixing that problem.

Gotcha: parameter ordering#

There’s one major flaw to the str() trick I’m using here: the order in which parameters are displayed in the string representation of call() may differ between Python versions. I had to undo this trick in one place I was using it (see here) as a result due to the following test failure:

1
E  At index 4 diff:
2
     "call().get_user_policy(PolicyName='policy-one', UserName='one')"
3
  != "call().get_user_policy(UserName='one', PolicyName='policy-one')"

Newsletter

Quick and dirty mock testing with mock_calls#

Gotcha: parameter ordering#