Fix for non-integer indexing in Numpy >= 1.12.0

[mperrin]

Issue by kmdouglass
Saturday Feb 18, 2017 at 09:54 GMT
Originally opened as mperrin/poppy#203

---

As of version 1.12.0, NumPy no longer implicitly casts non-integers to integers for array indexing and instead raises an IndexError. See the Numpy Release Notes for details.

A couple of lines in poppy_core.py and the test scripts required either explicit integer casting or integer division with the // operator for compatibility with NumPy >= 1.12.0.

The Poppy test suite is now passing on my machine with Python 3.5.2 and 2.7.13 and Numpy 1.12.0 with these fixes. You may want to add a build with Numpy 1.12.0 to Travis to verify that things work on your end.

---

kmdouglass included the following code: https://github.com/mperrin/poppy/pull/203/commits

[mperrin]

Comment by coveralls
Saturday Feb 18, 2017 at 10:07 GMT

---

Coverage remained the same at 64.564% when pulling 629488e on kmdouglass:noninteger_index_fix into 802a249 on mperrin:master.

[mperrin]

Comment by mperrin
Monday Feb 20, 2017 at 20:07 GMT

---

Very nice, thank you @kmdouglass. I have to admit I hadn't even realized numpy 1.12 was out yet. Much appreciated! I'll give this PR a day or so for code review & feedback from others here (not that it needs much) then pending that I expect this will be good to merge.

[mperrin]

Comment by kmdouglass
Tuesday Feb 21, 2017 at 05:54 GMT

---

You're welcome @mperrin . If there's something else that you need me to do, just let me know.

[mperrin]

Comment by kmdouglass
Wednesday Feb 22, 2017 at 15:39 GMT

---

Thanks for the feedback. If I understood everyone correctly I will amend this pull request by

1. removing the comments (I agree with @mperrin that they are a bit too verbose. NumPy makes it very clear why there is an indexing error if these are ever changed in the future);
2. amend the int(np.abs(np.floor((lx-lx_w)/2))) call to np.abs(lx-lx_w) // 2, leaving the np.abs call because the order of lx and lx_w is not known a priori;
3. remove the extraneous print statement (I'm sorry about this; I'll be more careful in the future.)

If this all seems correct then I'll make the changes. Thanks guys!

[mperrin]

Comment by mperrin
Wednesday Feb 22, 2017 at 15:56 GMT

---

That all sounds correct to me.

(And no need to apologize for an extra print statement - not a big deal. We all do this sort of thing; a big part of why Github is so great is that it makes this kind of code review easy and efficient. All code needs more than one pair of eyes on it!)

[mperrin]

Comment by coveralls
Wednesday Feb 22, 2017 at 17:21 GMT

---

Coverage remained the same at 64.564% when pulling a6ed74c on kmdouglass:noninteger_index_fix into 802a249 on mperrin:master.