Update Cython vs Numba.ipynb

[robertwb]

The primary difference in Numba vs. Cython here is that Cython refuses by default to infer fixed-size integer types for ints used in arithmetic (to avoid changing overflow behavior).

[elgehelge]

Wow. I did not know about that. This makes it much easier to use Cython. But can you tell me a little bit about the drawback of doing this. Why is it not the default?

[robertwb]

The goal is to be able to take your Python code, compile it with Cython,
and have it behave exactly the same. If you have (for example)

def squares(a, b):
for k in range(a, b):
print k*k

this will overflow whenever b > sqrt(sys.maxint). Even k+1 could overflow.

Granted, our heuristics are probably a bit too conservative. For
floats/doubles it's easy--we always infer.

On Thu, Jan 14, 2016 at 9:35 AM, Helge Munk Jacobsen <
notifications@github.com> wrote:

Wow. I did not know about that. This makes it much easier to use Cython.
But can you tell me a little bit about the drawback of doing this. Why is
it not the default?

—
Reply to this email directly or view it on GitHub
#1 (comment)
.