python - Custom location for .pypirc file

Question

Does setuptools allow for the .pypirc file to be specified in a custom location rather than $HOME/.pypirc? I'm setting up a jenkins job to publish to an internal repository, and want the .pypirc file to be inside the job's workspace.

Answer 1

You might want to check out twine. This tool allows you to specify a custom .pypirc file via the option --config-file

Answer 2

This is doable by overloading the default command classes used for communication with the PyPI or your custom repository server. This can be done in your setup.py script. Let the code do the talk first:

# this is your project's setup.py script

import os
from distutils.command.register import register as register_orig
from distutils.command.upload import upload as upload_orig

from setuptools import setup


class register(register_orig):

    def _get_rc_file(self):
        return os.path.join('.', '.pypirc')

class upload(upload_orig):

    def _get_rc_file(self):
        return os.path.join('.', '.pypirc')

setup(
    name='myproj',
    ...
    cmdclass={
        'register': register,
        'upload': upload,
    }
)

Now, if you call e.g.

$ python setup.py sdist register upload

the server credentials will be taken from .pypirc in the project directory instead of the default ~/.pypirc one.

So basically, all you need to do is to replace the register and upload commands with your own implementations. These implementations redefine the _get_rc_file method with your own logic. After that, don't forget to adjust the cmd_class argument in your setup function so your own command classes are used instead of the standard ones.

Of course, as both classes redefine the same method, you could prettify the code by e.g. writing a mixin class and reusing it etc etc. Also, if you are uploading the sphinx docs to PyPI via upload_docs command, you have to overload it the same way. Anyway, you should get the basic idea of command overloading from the snippet above.

Answer 3

In the last version of setuptools (version 29.0.1), the .pypirc file is loaded from the $HOME directory (or %USERPROFILE% on Windows).

This is done in the setuptools.package_index.PyPIConfig class. See the source code on GitHub:

class PyPIConfig(configparser.RawConfigParser):
    def __init__(self):
        """
        Load from ~/.pypirc
        """
        defaults = dict.fromkeys(['username', 'password', 'repository'], '')
        configparser.RawConfigParser.__init__(self, defaults)

        rc = os.path.join(os.path.expanduser('~'), '.pypirc')
        if os.path.exists(rc):
            self.read(rc)

So, the only solution I see is to temporary redefine the HOME environment variable before running the Jenkins job.