classification
Title: autospecced namedtuples should be truthy by default
Type: behavior Stage:
Components: Library (Lib) Versions: Python 3.4
process
Status: open Resolution:
Dependencies: Superseder:
Assigned To: michael.foord Nosy List: michael.foord, python-dev, skrisman, xtreak
Priority: normal Keywords: patch

Created on 2013-09-14 09:36 by skrisman, last changed 2020-07-28 05:11 by xtreak.

Files
File name Uploaded Description Edit
namedtuple_truthiness.patch skrisman, 2013-09-14 09:36 review
namedtuple_truthiness_2.patch skrisman, 2013-09-26 09:33 review
Messages (5)
msg197698 - (view) Author: Shawn Krisman (skrisman) Date: 2013-09-14 09:36
import mock
from collections import namedtuple

Foo = namedtuple('Foo', bar)
mock_foo = mock.create_autospec(Foo)

if mock_foo:
    print('the namedtuple is truthy')
else:
    print('the namedtuple is not truthy')


The expected behavior is that it should print "the namedtuple is truthy." Instead it prints "the namedtuple is not truthy." Almost all namedtuples are truthy, the exception being the nearly useless namedtuple with 0 fields. The problem stems from the fact that tuples have a __len__ defined which is what is used derive truthiness. MagicMocks define __len__ to be zero by default. Workarounds to the problem are very difficult because you cannot simply define __len__ to be a nonzero number and have the mock work correctly. 


In general MagicMock has defaults that encourage the mocks to be very truthy all around:

__bool__ -- True
__int__ -- 1
__complex__ -- 1j
__float__ -- 1.0
__index__ -- 1

So it was interesting to me to find out that __len__ was defined to be 0. The fix that I am proposing is to make 1 the new default for __len__. I believe this is a more useful default in general because an instance of a class with a __len__ attribute will likely be truthy far more often then not.


There are of course backwards compatibility issues to consider here, however I don't think many people are assuming this behavior. Certainly nobody in the python code base.
msg197889 - (view) Author: Michael Foord (michael.foord) * (Python committer) Date: 2013-09-16 10:33
The problem is that the change you're proposing is backwards incompatible. Code using MagicMock and objects with length will break.
msg197913 - (view) Author: Shawn Krisman (skrisman) Date: 2013-09-16 16:01
Yeah in my head I was thinking it would affect relatively few people who depended on the change, but it's definitely hard to prove that!

How about a change that special cases namedtuple?
msg198433 - (view) Author: Shawn Krisman (skrisman) Date: 2013-09-26 09:33
This fix is actually backwards compatible. This is a more powerful patch too because not only does it provide a better default for truthiness, but it also provides a better default for length. I also fixed a spelling mistake involving the word "calculate".
msg198486 - (view) Author: Michael Foord (michael.foord) * (Python committer) Date: 2013-09-27 15:53
I dislike adding namedtuple specific code to mock. How many different types should we special case here and in other places?
History
Date User Action Args
2020-07-28 05:11:43xtreaksetnosy: + xtreak
2013-09-27 15:53:51michael.foordsetmessages: + msg198486
2013-09-27 15:49:35eli.benderskysetmessages: - msg198484
2013-09-27 15:48:26python-devsetnosy: + python-dev
messages: + msg198484
2013-09-26 09:33:42skrismansetfiles: + namedtuple_truthiness_2.patch

messages: + msg198433
2013-09-16 16:01:54skrismansetmessages: + msg197913
2013-09-16 10:33:13michael.foordsetassignee: michael.foord
messages: + msg197889
versions: + Python 3.4, - Python 3.5
2013-09-14 09:36:45skrismancreate