weird dict problem, how can this even happen?

Joel Hedlund yohell at ifm.liu.se
Mon Dec 15 22:50:45 CET 2008


I'm having a very hard time explaining why this snippet *sometimes* 
raises KeyError:

snippet:
> print type(self.pool)
> for frag in self.pool.keys():
>     if frag is fragment_info:
>         print "the fragment_info *is* in the pool", hash(frag), hash(fragment_info), hash(frag) == hash(fragment_info), frag == fragment_info, frag in self.pool, frag in self.pool.keys()
> try:
>     renderer_index = self.pool.pop(fragment_info)
> except KeyError:
>     print "Glorious KeyError!"
>     for frag in self.pool.keys():
>         if frag is fragment_info:
>             print "the fragment_info *is* in the pool", hash(frag), hash(fragment_info), hash(frag) == hash(fragment_info), frag == fragment_info, frag in self.pool, frag in self.pool.keys()
>     raise


output:
> <type 'dict'>
> the fragment_info *is* in the pool 987212075 987212075 True True False True
> Glorious KeyError!
> the fragment_info *is* in the pool 987212075 987212075 True True False True
> Traceback (most recent call last):
>   File "/home/yohell/workspace/missy/core/gui.py", line 92, in process_job
>     renderer_index = self.pool.pop(fragment_info)
> KeyError: <core.gui.FragmentInfo object at 0x8fc906c>

This snippet is part of a much larger gtk program, and the problem only 
from time to time, predominantly when the cpu is under heavy load and 
this method gets called a lot. If I didn't know better I'd say it's a 
bug in python's dict implementation, but I do know better, so I know 
it's far more likely that I've made a mistake somewhere. I'll be damned 
if I can figure out what and where though. I've reproduced this bug (?) 
with python-2.5.2 on Ubuntu 8.10 and python-2.5.1 on WinXP.

I would very much like an explanation to this that does not involve 
threads, because I haven't made any that I'm aware of. I can't even 
understand how this could happen. How do I even debug this?

Please help, I feel like I've taken crazy pills here!
/Joel Hedlund



More information about the Python-list mailing list