packaging code with compiled libraries

[Oscar Benjamin]

On 5 October 2015 at 20:43, Tim <jtim.arnold at gmail.com> wrote:
>
> I have a package I want to share but have a question about packaging.
>
> Mostly the package is pure python code, but it also requires some binary libraries (*.so, *.dll, *.dylib).  I want to bundle these libs so users don't have to compile. The package will run on *nix/windows/mac platforms.
>
> Currently I handle this in setup.py. In the 'build' phase, I copy the platform-specific libs to a subdirectory called 'libs'.
>
>     class MyBuilder(build_py):
>         def run(self):
>             conditional logic for copying
>             appropriate library files to 'libs'
>             etc etc.
>             build_py.run()
>
> And that seems to work, but after reading more from the Python Packaging Authority, I wonder if that is the right way.  Should I be using wheels instead?
> I think my brain fried a little bit while going through the doc.

The idea of a wheel is that you want to distribute your code fully
precompiled to end users who will be able to install it without
needing any C compilers etc. Of course this requires you to supply
wheels for each platform of interest. If this is what you want to do
then yes absolutely use wheels. Note that if you have installed
setuptools and wheel and you use setuptools in your setup.py then
building a wheel is as simple as running "python setup.py bdist_wheel"
(once your setup.py is complete).

If the binary libraries in question are extension modules then you
should just declare them as such in your setup.py and
distutils/setuptools/wheel will take care of bundling them into the
wheel.

If the binary libraries are not extension modules and you are building
them separately (not using distutils) then you can declare them as
"datafiles" [1] so that they will be bundled into your wheel and
installed alongside your python code.

[1] https://packaging.python.org/en/latest/distributing/#package-data

--
Oscar