[Python-Dev] [Python-checkins] r41972 - python/branches/ssize_t/Objects/funcobject.c

["Martin v. Löwis"]

Neal Norwitz wrote:
> I often chose to use Py_ssize_t rather than int if it was capturing
> the result of a sequence length.  But often this is larger than what
> will be allowed (like in this case).  The problem then comes that we
> have cases everywhere if we leave things as ints.

This is also the strategy I have been following. The alternative would
be to add error handling at the places where truncation occurs (unless
you can *really* determine statically that there never ever will be
truncation). This is too painful, so it is IMO better to carry the wide
values throughout, until the place where the range check is carried out.

> Stuff like this:
>   len = (int) PyTuple_Size(co_consts);
> 
> How do we know that is safe or not?

Well, you know that LOAD_CONST only supports 2**31 constants, so
truncation to int is currently safe (I hope that the compiler detects
cases where somebody tries to create more than 2**16 constants).

Still, this is only the *current* state. Going further, it might be
that some of the restrictions are lifted someday, so we would have
to review all casts then to see whether they are still safe.

Therefore, I think having fewer casts is better.

Regards,
Martin