[tahoe-dev] [pycryptopp] #19: Segmentation fault in HashMultipleBlocks

[Nathan]

On Wed, Sep 9, 2009 at 8:50 AM, Zooko Wilcox-O'Hearn <zooko at zooko.com> wrote:
> On Wednesday,2009-09-09, at 9:43 , Nathan wrote:
>
>> Does Tahoe-LAFS intend to support pre-2.5 CPython?
>
> Yes, we support CPython >= 2.4.2 and < 3.0.
>
> Thank you for the debugging!
>
> Regards,
>
> Zooko
>

This means all wrapper code has to be hacked up in some manner (again,
I haven't discovered what the best definitions are) so that Py_ssize_t
is defined in a backwards compatible manner.  Actually I'm confused.
If we support python < 2.5, then where is Py_ssize_t defined for
wrapper code?  Is there a backport of the definition?

This also means that every cast from Py_ssize_t to something else
might be tricky!

Maybe what we need is a suite of compiled unit tests that try various
casts of Py_ssize_t to detect any subtle platform specific implicit
cast errors.  Actually, this seems like something the general python
community should already have.  (/me crosses fingers...)

Hm...  I'm going to add a "do more research on Py_ssize_t" ticket.