Newlines in the `description` field produce a malformed PKG-INFO

We discovered this accidentally by way of https://github.com/zopefoundation/zc.relation/issues/4#issuecomment-397532224: if you pass a string containing newlines to the description argument of setup(), setuptools will generate a malformed PKG-INFO.

To reproduce:

# setup.py
from setuptools import setup
setup(
    name='test-package',
    version='0.1',
    author='Blah Blah',
    author_email='[e-mail]',
    description='description\n\n',
    py_modules=['blah'],
)

(The contents of blah.py do not matter, but the file should exist.)

Run python setup.py sdist and the inspect test_package.egg-info/PKG-INFO. For me, with setuptools 39.1.0, it looks like this:

Metadata-Version: 1.0
Name: test-package
Version: 0.1
Summary: description


Home-page: UNKNOWN
Author: Blah Blah
Author-email: [e-mail]
License: UNKNOWN
Description: UNKNOWN
Platform: UNKNOWN

The extra newlines lead tools to treat the rest of the PKG-INFO as a long_description.

I would expect setuptools to complain about the newlines in the description field, or at least escape them properly (i.e. prepend whitespace, like it does for the long_description field).

[bittner]

I see the options as:

1. Escape newlines in the text so they aren’t interpreted as delimiters.
2. Raise an error whenever double newlines are found outside of long_description.
3. Silently s/\n+/\n/g for everything except long_description.

I think I prefer 1 as least disruptive, but I’m ambivalent between 2 and 3. 2 is more explicit, but 3 creates less “churn” (since this will just be auto-magically fixed).

We should fail hard for configurations that are invalid. There should be an error message that gives a clear indication on how to fix the problem. This is a developer API, and developers should deal with their broken code. We shouldn’t try to guess-fix broken stuff.

Do the fields/kwargs have an explicit specification of the data they expect? This should be enforced. The fields description and license obviously expect single-line data. Hence, when a newline is submitted in the data for those fields the generation of PKG-INFO should be aborted with a clear error message, e.g.

$ python3 setup.py sdist
ERROR: The "description" field requires a single line of text. Multi-line text was supplied.
Aborting.

[Mekk]

setuptools 51.3.0 landed and I immediately got some error report about broken (internal) builds.

Reporting error here is not a good idea, really.

Please, don’t go that way. Drop description altogether if you don’t like it and can’t fix it, but don’t break installations. Ecosystem is huge, libraries are plenty, automated builds are numerous…