[Python-Dev] 3.5 release schedule PEP

[Steve Dower]

> Paul Moore wrote:
> On 24 September 2014 14:16, Mike Miller <python-dev at mgmiller.net> wrote:
>> It has been a supported option for just shy of 15 years on 2.X...
>> most if not all the bugs (setuptools) were fixed a decade ago, and
>> right now thousands, if not millions of people are running it under
>> Program Files right now. I can vouch for several thousand because a
>> company I work for distributes Python and pip there for its customers
>> all around the world w/o issue.
>
> One thing that I presume would be an issue. Isn't Program Files protected in
> newer versions of Windows? I haven't tested this myself, so I may be wrong about
> this. So take the following with a pinch of salt.

It's protected very well in newer versions. You typically need to be an administrator AND have opted in to being able to modify system files without warning.

> Assuming so, that means that if Python is installed there, the standard "pip
> install XXX" would not work unless run in an elevated shell. We are currently
> trying to focus on a unified message for how users should install distributions
> from PyPI, by using pip install.
> I'm not sure it's a good idea to complicate that message yet by adding
> provisos about managing the system Python (which is the only one most Windows
> users will use).

This is my main concern. Until pip install --user is the default (or the fallback if there are no write permissions on the destination), a default that locks users out of the simplest PyPI experience is a genuine problem. Yes, users can elevate to run pip, but I'd prefer pip to use elevation if it has it and to use per-user if not.

There also isn't a great story for per-user Python installs on Windows, but that becomes fairly cheap with the installer rewrite.

> I know this is only the same situation as Unix users have, but Windows users
> don't have a distro offering packaged versions of PyPI modules.
> I also know we should be moving towards promoting --user, but I don't think
> we're quite ready for that yet. And my speculation doesn't compete with your
> real-life experience, certainly. But I would suggest carefully checking before
> making the switch.

A good reason to decide early on a change like this, or at least to promote it as an option in 3.5 and make it the default in 3.6.

Cheers,
Steve

> Paul