Parts of traceback is lost when reraising exceptions

[jodal]

This issue was reported by @adamcik at IRC after review of the Pykka source code.

At https://github.com/jodal/pykka/blob/master/pykka/gevent.py#L37 a gevent.Timeout exceptions is reraised wrapped in a pykka.Timeout exception. This reraise does not keep the traceback of the original exception.

At https://github.com/jodal/pykka/blob/master/pykka/future.py#L93 a exception thrown in a ThreadingActor is reraised in the thread of a consumer of a ThreadingFuture. This reraise does not keep the traceback of the original exception.

The exceptions should be reraised using the three-argument raise syntax:

raise exc_class, exc_instance, traceback

E.g.:

raise _Timeout, _Timeout(e), sys.exc_info()[2]

In cases where the exception is thrown in one thread and reraised in another thread, the first thread must make sure to pass on both the exception object and the traceback, found at sys.exc_info()[2].

[adamcik]

I would suggest changing the future.py code to pass around (value, exc_info) instead of just value. This seems more robust than assuming that any value that is an exception should be raised. Looking at other code bases that handle this type of re-raise issues there seem to be some gotchas here with respect to circular loops created by data in the traceback that you will probably have to avoid.