bpo-35186: Remove "built with" comment in setup.py upload

[pganssle]

platform.dist() is deprecated and slated for removal in Python 3.8. The upload command itself should also not be used to upload to PyPI, but while it continues to exist it should not use deprecated functions.

This is the first option I proposed in bpo-35186. A slightly more conservative option (option 2) is easy enough to implement:

diff --git a/Lib/distutils/command/upload.py b/Lib/distutils/command/upload.py
index 32dda359ba..56ba12b1c5 100644
--- a/Lib/distutils/command/upload.py
+++ b/Lib/distutils/command/upload.py
@@ -121,12 +121,9 @@ class upload(PyPIRCCommand):
             'requires': meta.get_requires(),
             'obsoletes': meta.get_obsoletes(),
             }
+
         comment = ''
-        if command == 'bdist_rpm':
-            dist, version, id = platform.dist()
-            if dist:
-                comment = 'built for %s %s' % (dist, version)
-        elif command == 'bdist_dumb':
+        if command in ('bdist_rpm', 'bdist_dumb'):
             comment = 'built for %s' % platform.platform(terse=1)
         data['comment'] = comment

Distutils maintainers let me know how you want to go. I'll hold off on the news entry until after we decide which way to go.

https://bugs.python.org/issue35186

[pganssle]

@di @dstufft Any chance you know if the "comment" field is used for anything in warehouse?

[di]

@pganssle Yes, the field is "used" in the sense that it's accepted and stored by PyPI. It's an optional, free-form field used to add comments to distributions. The comments are available in the API and appear as tooltips for the files in PyPI's management UI:

The format or contents of the comment here is not specifically used for anything in PyPI, though.

[pganssle]

I think as long as it's only used as a free-form comment it should be fine to remove it from distutils. Anyone who runs into this problem can and should use twine -c "built for $(some_platform)" or whatever.

At worst this is going to break people who are parsing the comment field from PyPI for whatever reason, and I suspect that's both exceedingly rare and not really something anyone is eager to support.

[dstufft]

Yea, what @di said. It has no semantic meaning, just a freeform whatever field.

Honestly we might even want to consider removing it, I'm not sure it really makes much sense in the modern world where almost all downloads from PyPI are automated tooling and not humans making project by project decisions on what file to download.

[pganssle]

Ok, it looks like we're leaning toward removing the comment, so I've added a news entry to that effect.

[pganssle]

@berkerpeksag @encukou @malemburg Are y'all the ones who are moving the deprecation of platform.dist() forward? I'm not sure who should review / merge this PR.

[merwok]

Does it make a difference to send empty string or omit comment altogether?

[pganssle]

The old version sent an empty string by default, so I kept that behavior, but I just checked what twine does, and it does seem to send '' even when comment=None (the default), so I think that distutils should do the same.

[encukou]

platform.dist() was removed because it was misleading – it misidentified distros, and there was no sustainable way to keep it up to date.
Efforts to use distro names instead of proper feature detection vs. keep the identification code up to date form a vicious cycle I explained in a distro bug.

I can't see how this comment is useful (esp. if it's added automatically), so I'd be for removing it.

[pganssle]

@encukou @berkerpeksag @merwok Anyone want to push the merge button?

Just a ping, let me know if there are any other concerns before merging.

[merwok]

@encukou Not sure if your last message means «+1 on this PR» or «please change something in the PR»

[encukou]

A +1. Thanks for the reminder!