Do not move stuff around in setup.py

[nils-werner]

Your current approach of moving stuff around in setup.py is very unclean:

Even without passing an actual action (install, sdist etc.) the script will copy stuff around. Also, you will not be able to build source packages for Windows when you are on any other operating system.

My solution: Always ship the DLLs and make the distinction in your actual module, not in setup.py.

[mgeier]

Should the dlls be part of the sdist tarball?
I tried it (on Linux) and they weren't.
They were included with bdist*, however.

[nils-werner]

Yes, they should be. Otherwise you will not be able to build it from PyPI. I have done something similar recently and will check what I did to fix it.

[nils-werner]

Regarding sndfile32bit.dll: I don't think it makes the code more readable so I would advise against it.

[nils-werner]

OK, I have implemented a solution that packages the DLLs in the sdist ZIP file. However I cannot test if Windows will be able to find the files in their new location. Can anybody check?

[bastibe]

Thank you for your contributions. Any simplification of the binary DLL handling is greatly appreciated. At the time, I merely went with the thing that worked, which is probably not the most elegant solution.

However, I wouldn't want to copy all binaries into all bdist packages.

Also, I don't think that we need to convert PySoundFile into a package. I like the fact that you can just import pysoundfile with just the file in your working directory.

[nils-werner]

Also, I don't think that we need to convert PySoundFile into a package. I like the fact that you can just import pysoundfile with just the file in your working directory.

I would go with whatever makes PIP installs more easy. I couldn't find a way to having a module and having package_data.

[bastibe]

Oh, I see.

[bastibe]

We'll try to get rid of the binaries altogether. We'll reevaluate this question once that process is done.

[mgeier]

I created an alternative to this PR: #98.

[mgeier]

I'm closing this, see #98 for the reasoning.