[TIP] tox and pypy and numpypy

[Albert-Jan Roskam]

----- Original Message -----

> From: Tres Seaver <tseaver at palladion.com>
> To: testing-in-python at lists.idyll.org
> Cc: 
> Sent: Friday, November 21, 2014 5:54 PM
> Subject: Re: [TIP] tox and pypy and numpypy
> 
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> On 11/21/2014 03:08 AM, Albert-Jan Roskam wrote:
> 
>>  I am using the latest version of Tox and I would like to include Pypy
>>  in my list of environments. One of the required packages is numpy. 
>>  Currently, tox tries to pip install the standard numpy (which fails), 
>>  not numpypy. What is the correct way to solve this? I just read about 
>>  'use_sitepackages=True', which might be an option.
> 
> You can set up the 'deps' for the environment differently.  E.g.:
> 
> 
> - ------------------------- %< --------------------------
> [tox]
> envlist = py26,py27,pypy
> 
> [testenv]
> deps =
>     nose
>     numpy
> commands =
>     nosetests
> 
> [testenv:pypy]
> deps =
>     nose
> 
>     numpypy
> commands =
>     nosetests
> - ------------------------- %< --------------------------
> 


hi,

Thanks, you put me on the right track. Numpypy and the normal numpy are both called 'numpy' though. 

I got it working with a separate 'dep', like you suggested, and by using the (experimental) 'install_command'


[tox]
envlist = pypy,py27,py33,py34,docs
skipsdist = True

[testenv]
deps = 
           setuptools
     nose
     gocept.testing
     numpy 
commands =
nosetests --with-xunit --xunit-file=junit-{envname}.xml --verbosity=3

[testenv:pypy]
deps = 
    setuptools
    nose
    gocept.testing
install_command = 
    pip install git+https://bitbucket.org/pypy/numpy.git {packages}


And it was worth the effort, because numpypy apparently does not 'upcast' arrays with mixed data types. This is with regular numpy (notice the ints become str, upcasting):
>>> numpy.array([['a', 1], ['b', 2]])
array([['a', '1'],
['b', '2']], 
dtype='|S1')

regards,
Albert-Jan