Fix traceback reporting for exceptions with __cause__ cycles.

[asottile]

Resolves #3804

[coveralls]

Coverage increased (+0.05%) to 92.555% when pulling 17644ff on asottile:cause_cycles into 64faa41 on pytest-dev:master.

[nicoddemus]

Ouch this formatting here is messy IMHO. How about this instead:

        out = "\n".join(line for line in tw.lines if isinstance(line, str))
        expected_out = textwrap.dedent(
            """\
            :13: in unreraise
                reraise()
            :10: in reraise
                raise Err() from e
            E   test_exc_chain_repr_cycle0.mod.Err

            During handling of the above exception, another exception occurred:
            :15: in unreraise
                raise e.__cause__
            :8: in reraise
                fail()
            :5: in fail
                return 0 / 0
            E   ZeroDivisionError: division by zero"""
        )
        assert out == expected_out

What do you think?

Otherwise the rest of the patch looks good, thanks!

[asottile]

ugh, I didn't even notice, black made this look way worse :(

[nicoddemus]

Thanks! 😁

[RonnyPfannschmidt]

i believe for exceptions its fine to use the direct object banking on the object identity hashing by default,
and that way this wouldn't incur the same cost as it does with id() on pypy

[asottile]

I didn't really want to make assumptions about exceptions since any user space exception could appear here, even one with an exceptionally bad __hash__ + __eq__. I considered using object.__hash__(e) instead of id(e) but I figure they're probably the same.

The "performance hit" in pypy is almost certainly negligible and this isn't a performance critical path (formatting failed test tracebacks).

[RonnyPfannschmidt]

thanks for the rebuttal