pynbody import error

[emapple]

The bug

I recently encountered the error below when trying to import pynbody, as did at least one other pynbody user. Similar issues have recently popped up in other packages (e.g., https://stackoverflow.com/questions/66193302/qiskit-error-numpy-ndarray-size-changed-may-indicate-binary-incompatibility, scikit-learn-contrib/hdbscan#457) since numpy released 1.20.0.

numpy.ndarray size changed, may indicate binary incompatibility. Expected 88 from C header, got 80 from PyObject

To Reproduce

I am using conda 4.9.2, and created an evironment with the environment.yml file below.
Steps to reproduce:

conda env create -f environment.yml
conda activate ufds
ipython
import pynbody

My setup in the described environment is below. I would have expected numpy to be version 1.20.0 given the new release, but it doesn't seem to be yet available in the default channels.

3.8.8 (default, Feb 24 2021, 21:46:12) 
[GCC 7.3.0]
Linux-2.6.32-696.13.2.el6.x86_64-x86_64-with-glibc2.10
numpy 1.19.2

environment.yml

name: ufds

dependencies:
    - python=3.8
    - ipython
    - jupyter
    - scipy
    - numpy
    - matplotlib
    - pandas
    - nb_conda
    - cython
    - pip:
        - "git+git://github.com/pynbody/pynbody.git"
    - seaborn
    - ipykernel
    - scikit-learn

[emapple]

After writing this out I realized I should try with the conda-forge channel, which does have numpy 1.20.0. Adding

channels:
    - conda-forge
    - defaults

to the environment file solves the issue for me (though this may be more of a workaround than a solution, depending on the channels you want to be using).

[MatthieuSchaller]

I am facing a similar issue when loading pynbody after having installed it via pip.

That is:

pip3 install pynbody
ipython3

And then:

Python 3.8.5 (default, Jan 27 2021, 15:41:15) 
Type 'copyright', 'credits' or 'license' for more information
IPython 7.13.0 -- An enhanced Interactive Python. Type '?' for help.

In [1]: import pynbody                                                                                                                                                                                                                                                                    
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-1-57137f64acd8> in <module>
----> 1 import pynbody

~/.local/lib/python3.8/site-packages/pynbody/__init__.py in <module>
     31 from .configuration import config, config_parser, logger
     32 
---> 33 from . import util, filt, array, family, snapshot
     34 from .snapshot import tipsy, gadget, gadgethdf, ramses, grafic, nchilada, ascii
     35 from . import analysis, halo, derived, bridge, gravity, sph, transformation

~/.local/lib/python3.8/site-packages/pynbody/util.py in <module>
     28 
     29 logger = logging.getLogger('pynbody.util')
---> 30 from ._util import *
     31 
     32 

__init__.pxd in init pynbody._util()

ValueError: numpy.ndarray has the wrong size, try recompiling. Expected 80, got 88

The issue goes away if I upgrade my numpy to 1.20.3 instead of the default 1.19.4 present on my platform.
Updating the pip dependencies might be the solution here.

[apontzen]

This is a built-in 'feature' of pip, I'm afraid, where it builds against a siloed version of numpy instead of the one you actually have installed.

Try passing --no-build-isolation to pip if you don't want to or can't update numpy.