On 28 March 2013 02:05, Nick Coghlan <ncogh...@gmail.com> wrote: > On Thu, Mar 28, 2013 at 11:43 AM, Donald Stufft <don...@stufft.io> wrote: >> I don't think you can, nor should you be able to, explicitly depend on >> something that is a VCS checkout. > > I find it more useful to think of the issue as whether or not you > allow publication of source tarballs to satisfy a dependency, or > *require* publication of a fully populated sdist. If you allow raw > source tarballs, then you effectively allow VCS checkouts as well. I > prefer requiring an explicit publication step, but we also need to > acknowledge that the installer ecosystem we're trying to replace > allows them, and some people are relying on that feature.
To give a real-life example of this issue, on Windows IPython depends on PyReadline. But the released version (1.7.x) of PyReadline is Python 2 only. So if you are using IPython on Python 3, you have to also depend on PyReadline from git. Now IPython doesn't declare a dependency on the VCS version (it just depends on "pyreadline"). And pyreadline is sufficiently stagnant that it hasn't declared anything much. But as an *end user* I have to make sure I force pip to install pyreadline from VCS if I want a working system. Paul. _______________________________________________ Distutils-SIG maillist - Distutils-SIG@python.org http://mail.python.org/mailman/listinfo/distutils-sig