➕ 9️⃣ 0️⃣ 0️⃣ 0️⃣ for this.

I just spent a good few hours reading various posts on Stack Overflow and experimenting unsuccessfully with my setup.py to try to get it to package my data files correctly. 🙅

Using MANIFEST.in and include_package_data to the exclusion of package_data and data_files is definitely more straightforward and got me what I wanted very quickly.

I would also link to the MANIFEST.in command syntax so people can find it easily from the guide.

Hmm, actually I'm not totally getting what I expect from MANIFEST.in and include_package_data.

This is what I have:

project/
    module/
        __init__.py
        project.py
        script.sh
    tests/
        test_errrrrthang.py
    random-file
    MANIFEST.in
    setup.py

If I include everything inside project/ in my MANIFEST.in, then it all gets packaged by python3 setup.py sdist into a .tar.gz. This is good. Without the MANIFEST.in, only the Python files in module/ get packaged by sdist; script.sh and everything outside of module/ are left out. This is also good.

Well, maybe not good, but at least so far I think this behavior is well-documented.

However, if I try to pip install that source distribution into a new venv, I only get the contents of module/. tests/ and random-file are nowhere to be found in the venv, even though they are in the source distribution. The same thing happens if I install from a wheel generated by bdist_wheel.

Is this expected? I'm running Python 3.5.1, pip 7.1.2, setuptools 18.2, and wheel 0.26.0.

Luckily in my case, I can rearrange stuff so that it all gets installed by putting random-file and even tests/ inside of module/.

But what is the advice we want to give people who don't necessarily want to do this?

"This will not be supported going forward" is totally an OK answer, by the way. I think the issue I'm raising here has already been discussed extensively over on pypa/pip#2874, and I think that's what @dstufft was getting at over there. Is that right?

And for the record @dstufft, the explanation of your rationale and vision in these two comments was very enlightening. Thank you for writing those up. As a new Python packaging user, I'm optimistic about the direction we're going in.