This may be intentional, but the behaviour changed between python2 and python3.  Want to make sure it's intentional as we're working (hacking) around this in pytest: https://github.com/pytest-dev/pytest/pull/4284

The actual impact on pytest is the use of `inspect.isclass`

Simplest reproduction:

class C(object):
    @property
    def __class__(self):
        raise AssertionError('fail')

isinstance(C(), type)


In python2.x:

$ python2 t.py
$

In python 3.x:

$ python3.7 t.py
Traceback (most recent call last):
  File "t.py", line 6, in <module>
    isinstance(C(), type)
  File "t.py", line 4, in __class__
    raise AssertionError('fail')
AssertionError: fail


In python2.x it appears there's code which intentionally avoids this case:

https://github.com/python/cpython/blob/ca079a3ea30098aff3197c559a0e32d42dda6d84/Objects/abstract.c#L2906-L2909

Mostly want to see if this is intentional or not, it does feel odd that `inspect.isclass` raises instead of returning `False` in this case, but it's unclear if it's a problem with `isclass` or `isinstance`

Arbitrary exceptions should not be silenced. Imagine that the user pressed Ctrl-C when the __class__ property implemented in Python was executed. The KeyboardInterrupt exception will be silenced in Python 2, and isinstance() will silently return incorrect result.

arbitrary, sure, but deriving from `Exception` maybe?

MemoryError and RecursionError are subclasses of Exception. And there are may be other exceptions that can be raised by virtually any code and depend on conditions out of your control.

I'm with Serhiy that exceptions should not be swallowed up unless there's a very good reason to, and in this instance there isn't if there's a bug in code as that low of a level as debugging that would be atrocious.