Re: NULL dereference when memory is tight

On Thu, Feb 24, 2011 at 9:52 AM, Federico Di Gregorio
<federico(dot)digregorio(at)dndg(dot)it> wrote:
> On 24/02/11 10:39, Brian Sutherland wrote:
>> Attached is a patch for another issue, though I'm not sure if calling
>> PyErr_NoMemory within libpq is sane.
>
> I won't call any Py* function without holding the GIL. That's why
> conn_notice_callback() uses its own data structure instead of a simple
> Python list.

I've fixed the potential access to uninitialized memory by simply
discarding the notice in case allocation fails. We could additionally
print the notice on stderr but I think the process is doomed anyway if
it just failed to allocate a struct of two pointers.

Brian, if you want to keep on playing with INFER, you should probably
pull the outstanding patches from my devel branch
<https://github.com/dvarrazzo/psycopg/> as in the last days I've
cleaned up several allocation glitches throughout all the library.

Thank you, cheers!

-- Danele