scipy: minpack2.cpython-38-darwin.so, 2): Symbol not found: _Py_SET_TYPE

macOS 10.15.5 Python 3.8.3 numpy 1.19.0 cython 0.29.20

create a new conda environment with Python 3.8.3
pip install numpy cython pytest pybind11
git clean -xdf
git checkout master
build scipy (4c7b25e4816f6246f30bdfbe5cd7d258f4f107e6) with python setup.py build
install scipy via pip install .

I then change out of the scipy directory and try to import scipy:

>>> import scipy.optimize as sciopt
sciopt.Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/andrew/miniconda3/envs/dev3/lib/python3.8/site-packages/scipy/optimize/__init__.py", line 387, in <module>
    from .optimize import *
  File "/Users/andrew/miniconda3/envs/dev3/lib/python3.8/site-packages/scipy/optimize/optimize.py", line 33, in <module>
    from .linesearch import (line_search_wolfe1, line_search_wolfe2,
  File "/Users/andrew/miniconda3/envs/dev3/lib/python3.8/site-packages/scipy/optimize/linesearch.py", line 16, in <module>
    from scipy.optimize import minpack2
ImportError: dlopen(/Users/andrew/miniconda3/envs/dev3/lib/python3.8/site-packages/scipy/optimize/minpack2.cpython-38-darwin.so, 2): Symbol not found: _Py_SET_TYPE
  Referenced from: /Users/andrew/miniconda3/envs/dev3/lib/python3.8/site-packages/scipy/optimize/minpack2.cpython-38-darwin.so
  Expected in: flat namespace
 in /Users/andrew/miniconda3/envs/dev3/lib/python3.8/site-packages/scipy/optimize/minpack2.cpython-38-darwin.so

pip installing the latest release works.

About this issue

Original URL
State: closed
Created 4 years ago
Comments: 21 (19 by maintainers)

Most upvoted comments

Looks like we’re all good here. I can’t think of an obvious place to document not to use

python setup.py build
pip install .

so I think we should close this?

rgommers on Feb 6, 2021

This just worked for me without error:

(base) $ conda create -n dev38 python=3.8
(dev38) $ conda activate dev38
(dev38) $ python --version
Python 3.8.5

(dev38) $ git clean -xdf
(dev38) $ pip install cython numpy pybind11

Collecting numpy
  Using cached numpy-1.19.1-cp38-cp38-macosx_10_9_x86_64.whl (15.3 MB)
Collecting cython
  Using cached Cython-0.29.21-cp38-cp38-macosx_10_9_x86_64.whl (1.9 MB)
Collecting pybind11
  Using cached pybind11-2.5.0-py2.py3-none-any.whl (296 kB)
Installing collected packages: numpy, cython, pybind11
Successfully installed cython-0.29.21 numpy-1.19.1 pybind11-2.5.0

(dev38) $ pip install .
Processing /Users/andrew/Documents/Andy/programming/scipy
  Installing build dependencies ... done
  Getting requirements to build wheel ... done
    Preparing wheel metadata ... done
Building wheels for collected packages: scipy
  Building wheel for scipy (PEP 517) ... done
  Created wheel for scipy: filename=scipy-1.6.0.dev0+96caa9d-cp38-cp38-macosx_10_9_x86_64.whl size=18749869 sha256=a60d81519364b81b2900dcaf130a96278794b55833ce566654eec67d48fbfedb
  Stored in directory: /private/var/folders/m8/gwtcncws12jf60xw5n6knwnw0000gn/T/pip-ephem-wheel-cache-zmin2im9/wheels/dd/e2/ce/5e25e40098d319817f14448037967682ebcd7efcbb8c9ded4c
Successfully built scipy
Installing collected packages: scipy
Successfully installed scipy-1.6.0.dev0+96caa9d


(dev38) $ cd ..
(dev38) $ python
Python 3.8.5 (default, Aug  5 2020, 03:39:04) 
[Clang 10.0.0 ] :: Anaconda, Inc. on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import numpy as np
>>> import scipy.stats
>>> exit()

python setup.py build; python setup.py develop

now also work.

What does not work is:

python setup.py build
pip install .

So @peterbell10, your comment is correct. What I didn’t know, and I’m not sure that it’s in the documentation, is that pip install . pins the numpy version for building. I was under the impression that it always used the current installed version (if it was recent enough). What is the recommended build/install route if one wants to use the installed version of numpy to build scipy?

@mattip, import numpy always worked.

andyfaff on Aug 10, 2020