This question tries to complement this one .
I have a python app which uses third party modules from PyPI. I want to package my app into a debian package, but don't know how to handle python dependencies which are not in debian/ubuntu repositories(packed as debian packages too)
Solution #1:
build the modules from PyPI right into my debian package.
Solution #2:
create debian packages for all PyPI modules I need using stdeb and add them to debian/ubuntu repositories.
I actually need a Solution #3 because I want to install PyPI dependencies when I install my debian package, preferably into a virtualenv!
What would be Solution #3? Do I need to tweak DEBIAN/preinst maintainer script?
Best Answer
I've spoken with some maintainers at the Debian IRC channel irc://irc.debian.org#debian-mentors, asking for the exact same thing, and the general consensus was:
Solution #1:
Integrating dependencies in your package by copying their source files over as a single codebase is very frowned upon. It would defeat the purpose of a packaging system that handles dependencies, updates, versioning, etc.
Solution #3:
Downloading non-debian packages on-the-fly when installing a binary (
.deb
) is a serious security risk, definitely a no-no. You wouldn't even be able to inspect the dependencies by extracting thedeb
, because they are downloaded and installed at install time. It's an approach that completely bypass the repositories system. No concerned user would be happy with a package that, behind the scenes (and asroot
, remember!), downloads additional untrusted software from untrusted sources. Yes, that would require fiddling withDEBIAN/postinst
(orpreinst
) and issuing awget
(or, in your case,pip install
), and that is the approach taken by Flash, Oracle Java, Steam and others. But that is proprietary, closed source software, so their security is none anyway.Solution #1.5:
You didn't mention it, but you could integrate the dependencies only at build time, ie, in the source package (the
.orig.tar.gz
,.debian.tar.gz
,.dsc
triad), by downloading from PyPi when creating the "binary" package (the.deb
). The instructions for thepip install
would go intodebian/rules
(notice the lowercasedebian
, as opposed to the binary package), and would be executed when you issuedebuild
ordpkg-buildpackage
.This is a middle-ground between #1 and #3. It mitigates (but not solve!) some of the issues of #3: at least you can inspect the final product, and the
.deb
would not require internet access at install time. All the risks and burdens are transferred from final user to the package maintainer. But, has the same problems as #1, as it bypasses most of the packaging system infrastructure. Afterall, handling dependencies (versions, updates, requirements, conflicts) is whydpkg
/apt
was created in the first place! :)Solution #2:
The One True Right Way™. You create debian packages for your dependencies, list them as requirements in your package, and ship all the
.debs
or source packages.From there, you have a number of options:
Submit the source packages, both your software and its dependencies, for inclusion to Debian. If accepted, they would be automatically available to all Debian users, including all derivatives like Ubuntu.
Upload the source packages to Launchpad, thus creating a PPA that any Ubuntu user (and its derivatives like Linux Mint) could easily add and install
Host your own debian repository in your website, that users from any Debian-based system could add to their
/etc/apt/sources.list.d
and use theapt
infrastructure to download, install and keep updated, (like the above!)Host the
.deb
files for direct download and install. Noapt
or automatic updates involved thought.As for how to package your PyPi dependencies (and your python software too!), there are a number of tools and references that make the process easy:
stdeb, as you mentioned. Oldie and goodie.
Pybuild, a new, amazing tool from Debian that supersedes
stdeb
.And many useful references:
Style Guide for Packaging Python Libraries
Debian Python Policy
Need help? Check those out:
Debian Mentors FAQ
Python Packaging Team