Extend numpy recipe to support python3

[krinnewitz]

This PR extends the existing numpy recipe to also support
python3crystax. Additionally, the existing recipe code is cleaned up a
bit and numpy is upgraded to version 1.15.1.

The existing patches were ported to the new numpy version.
To get numpy 1.15.1 working with python 2.7, an additional patch
was necessary.

This PR should resolve #882.

[AndreMiras]

Awesome work! It's looking good to me. But I didn't yet get a chance to actually try it on device.
Also the build is still passing for python2 + numpy.
By the way it would be great to add python3 + numpy support to the continuous integration. But that can be tackled in a different pull request.
Also at some point we should think about slowly dropping python2 support.

I approve the PR, but won't merge it unless somebody else confirms it's working on device. If nobody does I'll take time later to check it. Otherwise ping us again :)

[krinnewitz]

Thanks so far @AndreMiras. Since no one has picked up on this, yet - here is a very simple program that can be used as a quick starting point:

import sys
print ('sys.version is', sys.version)

import numpy as np
a = np.arange(15).reshape(3, 5)
print('numpy seems to work:\n', a)

[AndreMiras]

Thanks for the reminder, merged!

[AntoniRomera]

Hi, when i use this code, i got the next error:

x86_64-linux-gnu-gcc: error: unrecognized command line option '-mfloat-abi=softfp'
x86_64-linux-gnu-gcc: error: unrecognized command line option '-mfpu=vfp'
x86_64-linux-gnu-gcc: error: unrecognized command line option '-mthumb'
x86_64-linux-gnu-gcc: error: unrecognized command line option '-mfloat-abi=softfp'
x86_64-linux-gnu-gcc: error: unrecognized command line option '-mfpu=vfp'
x86_64-linux-gnu-gcc: error: unrecognized command line option '-mthumb'

[JohnBuckleyGitHub]

First of all thank you very much to krinnewitz for getting the Python 3 build going. It has been an open ticket for so long.

As it seems python3crystax is being dropped in favor of python3 (Google NDK). The numpy recipe only has python2 and python3crystax. Does the recipe use any functions that are specific to crystax?

[krinnewitz]

I think porting/extending the recipe for google NDK's python3 shouldn't be a problem.

[JohnBuckleyGitHub]

If somebody could take this on it would be super helpful, I have had a look and it is out of my depth. I am caught in that I need both SSL and Numpy. Crystax doesn't support SSL easily. So I have been using python2, but my AWS module boto is getting old and buggy. So a non-crystax python 3 would get me out of this squeeze.

…

On Mon, 31 Dec 2018 10:56 Kim Rinnewitz ***@***.*** wrote: I think porting/extending the recipe for google NDK's python3 shouldn't be a problem. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#1343 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACT71izdyVHLdU46sbvcGHnE8xktiswbks5u-e1WgaJpZM4WPzq4> .

[AndreMiras]

@johnbolia start a "WIP" pull request that try the fix and share your problem in that dedicated PR, we can also give support on Discord to help you building it