part of the assertion representation logic does try various things to tell more about the object

the isiterable helper unfortunately calls __iter__ instead of just verifying its existence

it only checks for type-error, which doesn't account for broken implementations of iter

we should decide whether we want to uphold the invariant that iter calls should only raise type errors, or whether we need to do something better

in any case the error needs to be better

Thanks, I actually did not think that there was some conventions about the type of exceptions being raised by special methods...

I'm working on a custom interpreter for a different language, so I simply had a different MyTypeError. Making a kind of hybrid error inheriting both type errors actually fix the issue.

CC @nicoddemus after reviewing i beleive we shoudl add 2 additional failure modes

a) types that wrap data but dont proviede __iter__ should diff a mu
ltiline repr
b) types that fail __iter__ should be considered differently

In [1]: 
   ...: 
   ...: class A:
   ...:     def __iter__(self):
   ...:         raise ValueError()
   ...: 
   ...:     def __eq__(self, o: object) -> bool:
   ...:         return self is o
   ...: 

In [2]: 

In [2]: a = A()

In [9]: pprint.pformat(a)
Out[9]: '<__main__.A object at 0x7fab6fda9d30>'

In [10]: from _pytest.assertion.util import _pformat_dispatch

In [11]: _pformat_dispatch(a)
Out[11]: '<__main__.A object at 0x7fab6fda9d30>'

In [12]: 

i believe a better error should be triggered

I agree with b), however I don't understand what you mean with a)... why use a multiline repr for types that don't have a __iter__ method?

@nicoddemus i have some helper pbjects around that just wrap around other sutff

also openapi types dont necessarily iterate but have a nested multiline repr

Hmm wait, I might be missing something, what do you mean by "multiline repr"? 🤔

@nicoddemus any type where repr is either multiline on its own or by using something some pprint.pformat

You mean calling __repr__().split() and see if there are more than one line, and treat that accordingly? I see.

Just use diff on it

Hi, guys.

When a test fails, Pytest fails to print repr() for objects in a list. Example output:

============================= test session starts ==============================
collecting ... collected 1 item
test_sample_users FAILED [100%]

sample_users = [<[KeyError('is_adult') raised in repr()] SampleUser object at 0x7fa9fad23a00>, <[KeyError('is_adult') raised in repr()] SampleUser object at 0x7fa9fad23be0>]

# the test traceback is not included, it does not matter

My SampleUser is a 'pydantic' class, and SampleUser.is_adult is a class property. I do not know why Pytest fails to call repr(), because I can repr() manually in my code and all works fine.

Versions: Python 3.10.1; pytest latest; pydantic master.

QUESTION: Is it related to this issue, or I need to create a new one with more details?