How do I make pip recompile Cython extensions when I have only changed setup.py when installing in editable mode. Currently it always skips the extensions. There are many questions related to this for distutils but I can't see any answers for pip.
To be clear, I have a setup.py like the following
from distutils.core import setup, Extension
from Cython.Distutils import build_ext
import numpy
ext_modules = []
# simulate_fast
ext_modules += [
Extension("adio.simulating.simulate_fast_c",
sources=["./adio/simulating/simulate_fast/simulate_fast_c.pyx",
"./adio/simulating/simulate_fast/c/simulate_fast.c",
"./adio/simulating/simulate_fast/c/matrix.c"],
include_dirs=[numpy.get_include()],
extra_compile_args=["-Ofast", "-ffast-math", "-march=native"],
language='c',
libraries=["gsl", "openblas"],
define_macros=[('FLOAT32', 1)]
)
]
setup(
name="adio",
packages=["adio"],
cmdclass={'build_ext': build_ext},
ext_modules=ext_modules
)
Now if I delete the ('FLOAT32', 1) in define_macros of course I would like the Cython extension to recompile.
However when I run
python3 -m pip install --editable -U . -v
I receive the following output as part of the output
running develop
running egg_info
writing adio.egg-info/PKG-INFO
writing dependency_links to adio.egg-info/dependency_links.txt
writing top-level names to adio.egg-info/top_level.txt
reading manifest file 'adio.egg-info/SOURCES.txt'
writing manifest file 'adio.egg-info/SOURCES.txt'
running build_ext
skipping './adio/simulating/simulate_fast/simulate_fast_c.c' Cython extension (up-to-date)
I have tried the -I and --force-reinstall flag with pip but it always skips the Cython extension. If I am not using editable mode then I can run
python3 -m pip install -U . -v
and this does recompile. How can I achieve the same thing when using the --editable flag.
Related
distutils ignores changes to setup.py when building an extension?
Related
I can customized the installing process in develop mode by setup.py like below:
from setuptools import setup
from setuptools.command import develop
class CustomDevelop(develop.develop, object):
"""
Class needed for "pip install -e ."
"""
def run(self):
super(CustomDevelop, self).run()
print("CustomDevelop=====================")
install_requires = ["numpy"]
setup(
name="example_package",
version="0.0.1",
packages=["example_package"],
cmdclass={'develop': CustomDevelop},
)
When running python3 -m pip install -vvv -e ., the output is
Running setup.py develop for example-package
Running command python setup.py develop
running develop
running egg_info
writing example_package.egg-info/PKG-INFO
writing dependency_links to example_package.egg-info/dependency_links.txt
writing top-level names to example_package.egg-info/top_level.txt
reading manifest file 'example_package.egg-info/SOURCES.txt'
writing manifest file 'example_package.egg-info/SOURCES.txt'
running build_ext
Creating /home/example_package/venv_example_package/lib/python3.8/site-packages/example-package.egg-link (link to .)
Adding example-package 0.0.1 to easy-install.pth file
Installed /home/lai/example_package
CustomDevelop=====================
However, it's not possible to do it with PEP518 which introduces pyproject.toml. Is there a workaround for this issue?
I just read the Cython Pure Python Mode documentation and I'm not sure if I understand it right. It sounds as if I could keep all my Python files as they are, add *.pxd files where I declare Cython types. In the setup.py, I still add
from setuptools import setup
from Cython.Build import cythonize
setup(
ext_modules = cythonize(
"A.py",
compiler_directives={'language_level' : "3"}
)
)
When I run python setup.py build_ext --inplace it actually builds the .so file.
What happens when I create the sdist / bdist, upload them to PyPI and a user does not have a matching platform? They will download the sdist, sure. I guess pip / setuptools will automatically try to compile the extension modules (A.py) and I guess if that works, it is fine. But what if cythonize fails? Will it still install the package and use the pure Python code?
I don't think so. I believe a failure in setup.py aborts installation completely.
You can try to declare an extension optional but there're reports that doesn't really work. Could be an issue with older setuptools.
When creating a Python package and uploading it to pypi, it will automatically install the requirements that are put in the setup.py file under install_requires, e.g.
from distutils.core import setup
setup(
name = 'a_package',
packages = ['a_package'],
install_requires=['another_package']
)
When the package has a cython extension (and .pyx files instead of .c/.cpp files), the setup.py file will need to import cython to create an installable extension, e.g.
from distutils.core import setup
from distutils.extension import Extension
from Cython.Distutils import build_ext
setup(
name = 'a_package',
packages = ['a_package'],
install_requires=['another_package'],
cmdclass = {'build_ext': build_ext},
ext_modules = [Extension('the_extension', sources=['a_file.pyx'])]
)
But since Cython is imported before executing the setup part, when trying to install this package through pip from source (rather than from a wheel) downloaded from pypi, it will fail to install due to not being able to import cython, as it has not reached the part with the requirements yet.
I’m wondering what can be done to ensure that a pip install of this package from pypi will install cython before it tries to import it. Adding a requirements.txt with cython does not seem to add automatic-install requirements for files downloaded from pypi.
Now, I realize it’s possible to just pip install cython before pip install thispackage, but I’m wondering if there’s a better fix that would allow to install the package along with cython directly from pypi when it’s not possible to run an additional command (without resorting to uploading the .c. files and ajusting the setup.py file to use them instead of the .pyx).
What you're describing is a "build time dependency", and this is precisely the use case "PEP 518 -- Specifying Minimum Build System Requirements for Python Projects" was created for.
You can specify cython as a build-time dependency by adding a pyproject.toml file like:
[build-system]
requires = ["cython"]
Then when installing your package with a modern version of pip (or another PEP 518 compatible installer), cython will be installed into the build environment before your setup.py script is run.
I am trying to debug a small cython project following the instructions from the official Cython page. but the command:
python-dbg setup.py build_ext --inplace
fails with the error below. I have seen responses to a similar issue here but I don't think it applies for me as I am running Cython installed via apt-get. Any help would be much appreciated.
Traceback (most recent call last):
File "build.py", line 4, in
from Cython.Build import cythonize
File "/usr/lib/python2.7/dist-packages/Cython/Build/init.py", line 1, in from Dependencies import cythonize
File "/usr/lib/python2.7/dist-packages/Cython/Build/Dependencies.py", line 51, in from Cython.Compiler.Main import Context, CompilationOptions, default_options
File "/usr/lib/python2.7/dist-packages/Cython/Compiler/Main.py", line 17, in from Scanning import PyrexScanner, FileSourceDescriptor
ImportError: /usr/lib/python2.7/dist-packages/Cython/Compiler/Scanning.so: undefined symbol: Py_InitModule4_64
[35101 refs]
I had the same problem.
In this article the cause of it is explained:
The problem [..] is the python-dbg binary however is incompatible with
modules not compiled with them
http://hustoknow.blogspot.de/2013/06/why-your-python-program-cant-start-when.html
My solution was to uninstall cython (which was installed via pip), clone the cython github repo and manually install it with python-dbg:
git clone git#github.com:cython/cython.git
cd cython
sudo python-dbg setup.py install
Afterwards I was able to python-dbg setup.py build_ext --inplace in my own code.
We have a local pypiserver and up to now we are happy with it.
But psycopg2 makes trouble. It wants pg_config to be installed. I don't see a reason why it should be installed on the mirror server. I understand that it needs to be on the client where I install psycopg2.
Is this a bug in psycopg2? Or do I use the wrong options the get the package to the mirror?
pypi#gray:~> pip install --no-deps --no-install -d packages psycopg2
Downloading/unpacking psycopg2
File was already downloaded packages/psycopg2-2.5.1.tar.gz
Running setup.py egg_info for package psycopg2
Error: pg_config executable not found.
Please add the directory containing pg_config to the PATH
or specify the full executable path with the option:
python setup.py build_ext --pg-config /path/to/pg_config build ...
or with the pg_config option in 'setup.cfg'.
Complete output from command python setup.py egg_info:
running egg_info
creating pip-egg-info/psycopg2.egg-info
writing pip-egg-info/psycopg2.egg-info/PKG-INFO
writing top-level names to pip-egg-info/psycopg2.egg-info/top_level.txt
writing dependency_links to pip-egg-info/psycopg2.egg-info/dependency_links.txt
writing manifest file 'pip-egg-info/psycopg2.egg-info/SOURCES.txt'
warning: manifest_maker: standard file '-c' not found
Error: pg_config executable not found.
Please add the directory containing pg_config to the PATH
or specify the full executable path with the option:
python setup.py build_ext --pg-config /path/to/pg_config build ...
or with the pg_config option in 'setup.cfg'.
----------------------------------------
Command python setup.py egg_info failed with error code 1 in /home/pypi/tmp/pip-build/psycopg2
Storing complete log in /home/pypi/.config/pip/pip.log
We use this pypiserver: https://pypi.python.org/pypi/pypiserver
I know this question is over a year old but the answer is pg_config is a linux-only module. You would need to install postgresql-devel to get it.