Closed thomasgoirand closed 8 years ago
Hi Thomas,
I don't have ready access to the Debian BTS, so I'm going to update here.
I don't 100% know what's going on here, but I think it is an issue with traceback2 which we started to exhibit after 55278e0e4a935f1d8c468a4d29a8133088806790 when we started to depend on it.
https://github.com/testing-cabal/traceback2/blob/master/traceback2/__init__.py#L449 assumes that if an exception has a __cause__
, then the cause has a traceback. Clearly this assumption does not always hold.
__cause__
is normally set when an exception is re-raised using the new Python 3 raise from
syntax. Python 3 correctly takes care of setting __traceback__
on __cause__
.
However, it turns out that Django has special code for setting __cause__
on database exceptions:
As per PEP-3134, a
__cause__
attribute is set with the original (underlying) database exception, allowing access to any additional information provided. (Note that this attribute is available under both Python 2 and Python 3, although PEP-3134 normally only applies to Python 3.)
(from exceptions.txt, some markup edits)
I would suggest that:
__cause__.__traceback__
to None__traceback__
in a getattr
Hope this helps, jml
traceback2 is a rolling backport of Python3's traceback module. Per https://docs.python.org/3/library/exceptions.html - "raise new_exc from original_exc
The expression following from must be an exception or None. It will be set as __cause__
on the raised exception."
The expectation in Python 3 is that __cause__
is either None or a valid exception object - which implies the __traceback__
attribute. The likely thing here is that Django is setting __cause__
(non-standard in Python2) without also setting the exceptions __traceback__
- which it needs to do by hand on Python2 (just as it's setting __cause__
by hand)
For reference I have opened a ticket on the Django side: https://code.djangoproject.com/ticket/25761
It will likely get fixed but it's not a high priority ticket so you might still want to rework the code to not make this assumption... if Django did the mistake, it's quite possible other projects did something similar.
Thanks; the place to do that would be in testing-cabal/traceback2 (which is the python stdlib code for traceback handling with only the necessary adjustments to work on Python2.
Hi.
Could you please follow-up on the opened Debian bug here? https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=802677
Cheers,
Thomas