Feature request: Add original exception object to pytest_exception_interact

[liiight]

I'd really like to have a generic exception handler, maybe via a hook. I use a test application and I'd like to raise custom exception and just catch them in one place like so:

def pytest_exception_handler(item, exc):
    if isinstance(exc, KeyError):
        pytest.skip('I knew this was gonna happen!)
    if isinstance(exc, MyException) and 'foo' in item.name:
        raise AnotherGenericException('That foo test is a problem!)

I searched before opening this issue, and couldn't find a way to do this, or even a discussion about this suggestion. If such exist, I apologize.

[pytestbot]

GitMate.io thinks possibly related issues are #56 (Feature: report exceptions in threads as errors), #553 (Handle exceptions in inspect.getsourcelines()), #1165 (Need to specially handle "historic" hooks), #1749 (Feature request: doctest report format as an option), and #3144 (Exit exception).

[avadhanij]

Can this not be achieved using the hook - pytest_exception_interact? I wish the actual exception object was given to this hook though. I would very much like to use isinstance(exc, KeyError) rather than call.excinfo.typename == 'KeyError'

[liiight]

oh, I apologize, I completely missed that hook. I'll check it out, thanks!

[liiight]

Well, silly me, this does exactly what I wanted, sorry.
However, like stated above, it would be very nice to actually pass the exception object instead of refferencing it via call.excinfo.

You could turn this:

def pytest_exception_interact(node, call, report):
    if call.excinfo.type is HTTPError and call.excinfo.value.response.status_code == 400:
        do_stuff()

To this:

def pytest_exception_interact(node, call, report, exc):
    if isinstace(exc, HTTPError) and exc.response.status_code == 400:
        do_stuff()

Which is arguably a lot nicer and pythonic.

Should I change the issue to reflect that?

On a sort of unrelated note, and I apologize if I'm hijacking (my own) issue, is it possible to skip a test that failed during the setup phase? Using pytest.skip() in the above scenario when the exception occurs in the setup phase results in an INTERNALERROR

[RonnyPfannschmidt]

@liiight perhaps - the hook could be improved, and we'd like to get rid of callinfo in external apis in future ^^

[liiight]

I updated the title. I'll leave the description as is for prosperity's sake.
I'd like to take a stab into doing this, about time I give something back to this awesome project

[liiight]

Ok, started to take a look into this. As I see it there are at least two approaches to this:

1. The "ideal" design, which is to change the signature of the hook and remove

def pytest_exception_interact(node, report, exc):
    if isinstace(exc, HTTPError) and exc.response.status_code == 400:
        do_stuff()

The downside here is that it's a breaking change, the upside is that it's the nicest way IMO.

2. Add the exception to the call object, set a deprecation notice in the docs for callinfo:

def pytest_exception_interact(node, call, report):
    if isinstace(call.exc, HTTPError) and call.exc.response.status_code == 400:
        do_stuff()

I'm probably missing some broader aspect here though, as I'm not exactly sure where else callinfo is used.