Oh I'm not sure. As long as it's up to date I'm indifferent.

I think that I would prefer the pypi URL.

Let's see what others say.
Cc @ogrisel @amueller

@NickGoumas, please remember to mention the associated issue in your PR description (I edited the PR for you this time). The best is to use "Fix #issueNumber", this way the issue gets closed automatically when the PR is merged.

@GaelVaroquaux look at the associated issue. The link appears on PyPI so that would point to the page you are already in. To be honest I am not sure what DOWNLOAD_URL is supposed to be.

As I said in the associated issue, it would be great to get rid of all the outdated sourceforge links across the whole code base.

For the record download_url is not used in PyPA setup.py so maybe we can remove it.

fwiw from when i edited the releases link in the README (#7637 (comment))

I think pypi should be the ultimate source for releases, not github. - @amueller

if we decide github should be the source for releases, maybe edit the readme as well?

FWIW if you do decide GitHub, would recommend creating your own tarball and adding it to the published tags and have the URL point to that. The reason I suggest this is we learned at conda-forge last summer that GitHub generates their tarballs on the fly. So if the process for generating them changes, so does the checksum. This has happened a couple times to us in the last year and started occurring to us this year too. 😞 Publishing your own tarballs on GitHub would avoid this issue. Alternatively using PyPI avoids this issue as well.

I am in favour of using PyPI and removing the DOWNLOAD_URL from setup.py unless someone sees a good reason to keep it there.

@NickGoumas are you still working on this?

@lesteve Is the consensus we just want to remove the DOWNLOAD_URL ?

@lesteve Is the consensus we just want to remove the DOWNLOAD_URL ?

Nobody strongly complained, so I guess so.

Ideally:

• somebody would double-check that a setup.py without DOWNLOAD_URL is perfectly valid, for example from the distutils and/or setuptools documentation
• it would be great to remove all the links pointing from sourceforge as it was mentioned in the associated issue

In the documentation for setup.py configuration its written that download_url .For conforming this I saw this blog-http://peterdowns.com/posts/first-time-with-pypi.html and also the official disutil documentation - https://setuptools.readthedocs.io/en/latest/setuptools.html. The mechanism they say is that it is one of primary parameters.
Usage:(from setuptool page)
download_url and url:
These become links on your project’s PyPI page. EasyInstall will examine them to see if they link to a package (“primary links”), or whether they are HTML pages. If they’re HTML pages, EasyInstall scans all HREF’s on the page for primary links.

When making a package and uploading it to pypi in the setup.py of package we must use the parameter ###The download_url is a link to a hosted file with your repository's code. Github will host this for you, but only if you create a git tag

tl;dr:
If you want it to download the tarball from github or sourceforge use download_url otherwise not needed.I saw the last official reference.https://pythonhosted.org/an_example_pypi_project/setuptools.html.
So please tell I will be happy to provide a pr based on your decision.

Thanks for taking a stab at this one. Because things have changed quite a bit in the past few years with python packaging, I think the most convincing thing to do would be to:

• remove download_url from setup.py. Rename the package name in setup.py to something like my-scikit-learn-test while you are doing your tests.
• test that you can upload the package to testpypi.python.org (I think -r option in python setup.py may be enough but I haven't tested it)
• test that you can pip install from testpypi.python.org (with -i option in pip install)

@lesteve I have tested we can upload successfully but when we install I am getting error:

Error generated:

@shubham0704 did you install numpy?

Oh my bad. But is there a provision to download required dependencies if not detected in the system. Like I used to make a requirements file while hosting a website -@nelson-liu.(I assumed that way)
I put this inside a virtual-env.
I will try installing all the dependencies and then try reporting a problem if any.Thanks

No errors after installing scipy and numpy: