segmentation fault when unpickling an object in another Python instance

[Raik Gruenberg]

Michael Hudson wrote:
> Raik Gruenberg <graik at web.de> writes:
> 
> 
>>It's not cPickle specific - I got the same segmentation fault when
>>using normal pickle.
>>
>>Getting a minimum example will take some time and it can get quite
>>long. I can switch off the error by deleting a line in the constructor
>>of the pickled object...
>>
>>Summary so far, I can not unpickle my object if:
>>- The constructor of this object called (before pickling) a certain
>>method of an internally referenced object
>>- this internal object encapsulates an Numeric array
>>- the certain method deletes parts of this array (via Numeric.compress )
>>
>>Rings a bell for someone??
> 
> 
> I'd like to apply my SEP[1] field and say this is likely to be a bug
> in Numeric.  Not sure though.  Can you try just pickling & unpickling
> a Numeric.array that has had bits chopped out of it with .compress()?

You are right, sort of. The problem is narrowed down: It, basically, 
seems impossible to unpickle a Numeric array of dictionaries (objects in 
general?) in another interpreter instance. I, nevertheless, submitted 
the bug report to the (more active) Python not the Numpy project. You 
shouldn't get a segfault, after all...

Well, I guess you could forward it to the Numpy team on grounds of [1] :)

Raik
> 
> Cheers,
> M.
> [1] Someone Else's Problem: I'm a developer on the Python project, but
>     not NumPy :)
>