Hello,

I've made a small change to try this suggestion (sending to the logger the original exception's exc_info, while keeping the ExceptionInfo object for internal purposes), and the change is effective, since for example the Django error-reporting engine is able to retrieve the original exception's information such as the traceback local variables.

def _log_error(self, task, req, einfo):
    ...

    logger.log(policy.severity, policy.format.strip(), context,
               exc_info=sys.exc_info() if policy.traceback else None,  # CHANGED HERE: exc_info=sys.exc_info(), instead of ExceptionInfo's exc_info
               extra={'data': context})

The test suites pass, and as reported in the CONTRIBUTING I've checked that the coverage does not decrease.
I have some difficulties, though, on coming up with a new test for this change, since it involves only what the logger receives as exc_info. Is a test for this needed, considering that the Celery internals still keep using the ExceptionInfo object, and the current tests pass? In case it's needed, do you have any suggestions on how could this be effectively tested?

Thanks