Thanks @johnthagen !

I'm no expert on setup tools, so I guess the question is if this is canonical or if pip-license should have been looking in the subdirectories?

@jklymak This brings up a good point, in the source dist (.tar.gz file), the LICENSE folder/files are already included because they have been added to MANIFEST.in.

But they are not included in any of the wheel distributions. pip-licenses looks at the current virtual environment to find the LICENSE files, so if they are not included in the wheel there is no way for pip-licenses to find them.

The subfolder is an important note though. I'll check if pip-licenses in its current form can handle a folder of LICENSEs if they are included.

Yeah it was a comment from complete ignorance. It just seems we can't be the only project that has a sub directory? Not at all objecting to this PR though.

From what I can tell, after matplotlib is installed, currently the LICENSEs don't end up in site-packages/matplotlib-3.3.1-py37.egg where they end up for other packages.

I think that setuptools/wheel have some default smarts to pull LICENSE* files into there when they are in the top directory. Since these are nested inside a folder, they don't get dropped in there by default. This PR should fix that.

I'm going to try to test the branch out with the sdist.

Well, it looks like perhaps license_files doesn't affect source dists: https://stackoverflow.com/questions/9977889/how-to-include-license-file-in-setup-py-script#comment104408651_48691876

So while the LICENSE files are included in the .tar.gz file, they aren't installed into site-packages/matplotlib-3.3.1-py37.dist-info (because there must be some magic in setuptools that looks for LICENSE* in the root folder).

I think this should work for wheels though, which is the default install most users will use.

Another option is to put multiple LICENSE files in the root of the project. opencv-python does this and they append all of the third party licenses together into a single file: https://github.com/skvark/opencv-python

I am 👎 on moving the license files around or merging them into one file. There are already a bunch of tools out there that expect them to be where they are. If setuptools / wheel can not cope with this we should push back on that being an upstream bug.

So after some further investigation, I've now realized that this will need to be fixed via setup.cfg:

[metadata]
license_files = LICENSE/*

This was the only way that actually got the LICENSE files into the wheel. setuptools and wheel seem to have a lot of automatic support for looking for LICENSE files in the root of the project, which had skewed my understanding when solving this in the past.

The important reading:

• https://wheel.readthedocs.io/en/stable/user_guide.html#including-license-files-in-the-generated-wheel-file
• Automatically include LICENSE files in source distribution pypa/setuptools#1636
• https://setuptools.readthedocs.io/en/latest/history.html#v42-0-0

I moved this PR to a Draft since I don't want to merge this as is as it doesn't actually solve the problem (sorry for not realizing this earlier).

matplotlib has a setup.cfg.template, is this used for the official builds pushed to PyPI? i.e. if we add the following lines to the template will that get used in the build?

I tested locally on macOS and with the setup.cfg changes in place in a file named setup.cfg, I can confirm the licenses get pulled into the wheel correctly.

In my view the fact that matplotlib reuses setup.cfg for its own purposes, even though it's really here for setuptools (distutils), is a minor but real annoyance (see e.g. the hacky del sdist.sdist.make_release_tree at the top of setup.py!) so I would be in favor of moving it to a separate mplsetup.cfg (or even mplsetupcfg.py, hehe).

Can you also add a line to https://github.com/matplotlib/matplotlib/blob/master/.github/workflows/cibuildwheel.yml to copy the template over when we build the wheels?

Can you also add a line to https://github.com/matplotlib/matplotlib/blob/master/.github/workflows/cibuildwheel.yml to copy the template over when we build the wheels?

@tacaswell Done. First time working with GitHub actions though. Is there a way to look at the built wheels for this PR so I can validate the LICENSE files made it in?

I don't think we build wheels on every PR, but it is probably worth adding a job to travis or azure to build the wheel and check if the licences are in it to make sure we don't break this in the future. You can also add a check after we run the ciwheel build in the GH actions to not publish wheels without the license.

Wheels are currently built on master, v\d+.\d+.x and tags. They are not restricted by fork though, so you can simulate it by pushing to your own master branch. Just make sure to reset it afterwards, or you'll have the extra commits haunting you.

For reference, the LICENSE files will be in:

matplotlib-<VERSION>-<PLATFORM>.whl
  matplotlib-<VERSION>.dist-info
    LICENSE*
    ...

The simplest way I've found to check if the LICENSE files are being added to the wheel is:

$ python setup.py bdist_wheel | grep 'adding license file "LICENSE/LICENSE"'

This will return success/0 if the LICENSE files were added and failure/1 if not. This method may be a little brittle though as it relies on the stdout of wheel.

Here is a Python script that will check if the LICENSE files are in the .whl files in dist. Is there a preference on which solution is used and if is so, where the script should live, etc.?

#!/usr/bin/env python3

"""Check that all .whl files in the dist folder have the correct LICENSE files included."""

from pathlib import Path
import sys
import zipfile

EXIT_SUCCESS = 0
EXIT_FAILURE = 1

LICENSE_FILE_NAMES = [
    'LICENSE',
    'LICENSE_AMSFONTS',
    'LICENSE_BAKOMA',
    'LICENSE_CARLOGO.txt',
    'LICENSE_COLORBREWER',
    'LICENSE_QT4_EDITOR',
    'LICENSE_STIX',
    'LICENSE_YORICK',
]


def main() -> int:
    dist_dir = Path(__file__).parent / 'dist'
    for wheel in dist_dir.glob('*.whl'):
        with zipfile.ZipFile(wheel) as f:
            license_file_names = [Path(path).name for path in sorted(f.namelist())
                                  if '.dist-info/LICENSE' in path]
            if license_file_names != LICENSE_FILE_NAMES:
                return EXIT_FAILURE
    return EXIT_SUCCESS


if __name__ == '__main__':
    sys.exit(main())

That script could be put in ci/; the LICENSE_FILE_NAMES could be generated from the same glob, and I'm not sure there's any need to have a main function.

@QuLogic I have updated the files as requested. GitHub Actions now has a step that validates that LICENSE files has been included in all built wheels.

I applied @QuLogic's latest suggestion, good catch. I believe this PR is ready for final review/merge.