On Oct 21, 2009, at 6:15 PM, Matthew Brett wrote:
> Hello,
>
> Please forgive my ignorance, but I am running into a problem that
> reveals my ignorance of reference counting and Cython in general.
>
> I am running code that creates a dictionary, fills it with lists of
> tuples created from C integers, and returns the dictionary, like this:
>
> def afunc():
> cdef int i=1
> cdef int j=2
> cdef int k=3
> cdef int m=4
> cdef int n=5
>
> d = {}
> val1 = (i, j)
> val2 = (k, m)
> d[m] = [val1, val2]
> return d
>
>
> The code above does indeed return the dictionary I was expecting:
>
> {5: [(1, 2), (3, 4)]}
>
> however, in a more complex routine, I am doing a similar thing, but
> returning many thousands of lists of tuples, and the routine grinds to
> a halt.
Meaning it doesn't terminate, runs out of memory, or segfaults?
> If I only return lists of integers, as in:
>
> d[m] = [i, j]
>
> The code runs at good speed and returns the expected result.
>
> Looking at the generated C code, I wonder if I am running into a
> reference counting problem, because the tuple I'm creating seems to be
> decref'ed.
>
> I've tried adding Py_INCREFs to the list, tuple, and python copies of
> the integers, but this made no difference to the larger routine.
>
> I realize I'm not understanding on-the-fly python object creation /
> destruction in Cython, and was wondering where I should go to look to
> understand it better. Is there anything obvious I have missed that
> would explain my problem?
Well, it's hard to say without seeing the code. I would try running it
with the reference nanny turned on and see if that turns up anything
suspicious. Also, try to whittle down your issue as small as possible
of an example that still fails and post it.
- Robert
_______________________________________________
Cython-dev mailing list
[email protected]
http://codespeak.net/mailman/listinfo/cython-dev
