[Twisted-Python] inlineCallbacks eating unhandled exceptions?

[Glyph Lefkowitz]

> On Sep 24, 2015, at 9:30 PM, Timothy Allen <Timothy.Allen at optiver.com.au> wrote:
> 
> The other day I was experimenting with the inlineCallbacks decorator in Twisted 15.4, and occasionally when I'd typo a variable name or similar, I'd  get an "unhandled error" in the log without a traceback, which was rather unhelpful. Eventually, I managed to produce the following test-case:
> 
>    https://gist.github.com/OptiverTimAll/43263a7a4cf5a298e8f5
> 
> If I run it as-is, I get the following unhelpful output:
> 
>    CRITICAL:twisted:Unhandled error in Deferred:
>    CRITICAL:twisted:
> 
> If I comment out the line that redirects Twisted's logging to the standard library's logging system, I get much more helpful output:
> 
>    Unhandled error in Deferred:
> 
> 
>    Traceback (most recent call last):
>      File "./ic-test.py", line 20, in <module>
>        bogus_callback()
>      File "/home/timall/.virtualenvs/txfuse/lib/python2.7/site-packages/twisted/internet/defer.py", line 1274, in unwindGenerator
>        return _inlineCallbacks(None, gen, Deferred())
>    --- <exception caught here> ---
>      File "/home/timall/.virtualenvs/txfuse/lib/python2.7/site-packages/twisted/internet/defer.py", line 1128, in _inlineCallbacks
>        result = g.send(result)
>      File "./ic-test.py", line 17, in bogus_callback
>        yield does_not_exist()
>    exceptions.NameError: global name 'does_not_exist' is not defined
> 
> I expected that changing the destination of log messages wouldn't change their content, but apparently I was wrong. I got the incantation for re-routing Twisted's logging system into the Python one by stitching together code-snippets from http://twistedmatrix.com/documents/current/core/howto/logger.html#starting-the-global-log-publisher Is there some better way, or some other thing I should be doing?

Your setup sounds basically correct, and I think you've just discovered a bug here.  Could you report it in the tracker?  I haven't fully diagnosed it yet, but in attempting to do so, I've found a few other odd edge cases in traceback printing.

-glyph